JP2009140454A - Data processor, data processing method, and program - Google Patents

Abstract

<P>PROBLEM TO BE SOLVED: To enable a user to perform a desired operation in advance. <P>SOLUTION: A state data acquiring part 101 acquires state data being time sequential data expressing the states. An operation data acquiring part 102 acquires desired operation data being the time sequential data corresponding to the operation desired by the user. A prediction learning part 103 learns the dynamics of the state data and the desired operation data. A predicting part 105 obtains the prediction value of the desired operation data with the state data as an input, on the basis of the dynamics. An operation data output part 106 outputs the prediction value of the desired operation data. The present invention is applied to an electronic apparatus to be operated by the user, such as a PC, a TV, and a game machine. <P>COPYRIGHT: (C)2009,JPO&INPIT

Description

  The present invention relates to a data processing device, a data processing method, and a program, and more particularly, to a data processing device, a data processing method, and a program that enable, for example, an operation that a user is about to perform in advance.

  For example, in a user interface of an electronic device such as a PC (Personal Computer) or a TV (TV receiver), the user can operate a mouse or a remote commander even if the user has performed in the past. It was necessary to repeat the same as in the past.

  A user interface using the user's line of sight has been proposed (see, for example, Non-Patent Document 1).

  According to such a user interface, the user can operate the computer only by moving the line of sight without operating the mouse or the remote commander.

Takehiko Ohno, “Interface Using Eyes,” Information Processing Vol. 44, No. 7, July 2003

  The previously proposed user interface using gaze detects a “stray” state or estimates word comprehension from a specific gaze movement pattern common to various users. In other words, the operation of the electronic device is fixedly associated with the estimation result or the like to support the user's work.

  Therefore, with the previously proposed user interface, for example, it is difficult for each user to recognize the movement of the line of sight unique to the user and perform the operation that the user is trying to perform according to the movement. It was.

  The present invention has been made in view of such a situation, and makes it possible to perform in advance a user's operation, that is, an operation desired by the user.

  A data processing apparatus or program according to an aspect of the present invention is a data processing apparatus that processes time-series data, and a user desires a situation data acquisition unit that acquires situation data that is time-series data representing a situation. Operation data acquisition means for acquiring desired operation data that is time-series data corresponding to an operation, learning means for learning the dynamics of the situation data and desired operation data, and the situation data as input based on the dynamics, A program that causes a computer to function as a data processing apparatus or a data processing apparatus that includes a predicting unit that calculates a predicted value of desired operation data and an output unit that outputs the predicted value of desired operation data.

  A data processing method according to one aspect of the present invention is a data processing method of a data processing apparatus that processes time-series data, acquires situation data that is time-series data representing a situation, and corresponds to an operation desired by a user. Obtaining desired operation data as time series data, learning dynamics of the situation data and desired operation data, obtaining the predicted value of the desired operation data based on the dynamics, using the situation data as input, and obtaining the desired operation A data processing method including a step of outputting a predicted value of data.

  In one aspect of the present invention, situation data that is time-series data representing a situation and desired operation data that is time-series data corresponding to an operation desired by the user are acquired, and the dynamics of the situation data and the desired operation data are acquired. Learning is done. Then, based on the dynamics, with the situation data as an input, a predicted value of the desired operation data is obtained, and a predicted value of the desired operation data is output.

  Note that the data processing device may be an independent device or an internal block constituting one device.

  Further, the program can be provided by being transmitted via a transmission medium or by being recorded on a recording medium.

  According to an aspect of the present invention, an operation desired by a user can be performed in advance.

  Embodiments of the present invention will be described below. Correspondences between the constituent elements of the present invention and the embodiments described in the specification or the drawings are exemplified as follows. This description is intended to confirm that the embodiments supporting the present invention are described in the specification or the drawings. Therefore, even if there is an embodiment that is described in the specification or the drawings but is not described here as an embodiment that corresponds to the constituent elements of the present invention, that is not the case. It does not mean that the form does not correspond to the constituent requirements. Conversely, even if an embodiment is described herein as corresponding to a configuration requirement, that means that the embodiment does not correspond to a configuration requirement other than the configuration requirement. It's not something to do.

A data processing apparatus or program according to one aspect of the present invention is provided.
A data processing device that processes time-series data (for example, the data processing device of FIG. 1),
Situation data acquisition means (for example, the situation data acquisition unit 101 in FIG. 1) for acquiring situation data that is time-series data representing the situation;
Operation data acquisition means (for example, the operation data acquisition unit 102 in FIG. 1) for acquiring desired operation data that is time-series data corresponding to an operation desired by the user;
Learning means for learning the dynamics of the situation data and desired operation data (for example, the predictive learning unit 103 in FIG. 1);
Based on the dynamics, prediction means (for example, the prediction unit 105 in FIG. 1) that obtains a predicted value of the desired operation data using the situation data as an input;
A data processing apparatus including an output unit (for example, the operation data output unit 106 in FIG. 1) that outputs a predicted value of the desired operation data, or a program that causes a computer to function as the data processing apparatus.

A data processing method according to one aspect of the present invention includes:
A data processing method of a data processing apparatus for processing time series data,
While acquiring the situation data which is the time series data representing the situation, the user obtains the desired operation data which is the time series data corresponding to the operation desired by the user (for example, steps S101 and S102 in FIG. 3).
Learning the dynamics of the situation data and desired operation data (for example, step S104 in FIG. 3),
Based on the dynamics, the situation data is used as an input to obtain a predicted value of the desired operation data (for example, step S113 in FIG. 4).
The predicted value of the desired operation data is output (for example, step S114 in FIG. 4).
A data processing method including steps.

  Embodiments of the present invention will be described below with reference to the drawings.

  FIG. 1 is a block diagram showing a configuration example of an embodiment of a data processing apparatus to which the present invention is applied.

  The data processing apparatus in FIG. 1 constitutes a part of an electronic device operated by a user, such as a PC, a TV, or the like.

  In FIG. 1, the data processing apparatus includes a situation data acquisition unit 101, an operation data acquisition unit 102, a prediction learning unit 103, a dynamics learning model storage unit 104, a prediction unit 105, and an operation data output unit 106.

The situation data acquisition unit 101 acquires situation data s ⁱ _t , s ⁱ _{t + 1} ,..., S ⁱ _{t + T} that are time-series data representing the situation, and sends them to the prediction learning unit 103 and the prediction unit 105. Supply.

Here, s ⁱ _t represents a sample value at time t of the i-th (i = 1, 2,..., I) type of situation data.

  The situation data acquisition unit 101 stores situation data representing various situations such as a mouse cursor, an icon, a window, a button displayed on a GUI (Graphical User Interface), and a user's line of sight on a PC or TV. get.

The operation data acquisition unit 102 acquires desired operation data a ^j _t , a ^j _{t + 1} ,..., A ^j _{t + T} that are time-series data corresponding to the operation desired by the user, and the prediction learning unit 103. To supply.

Here, a ^j _t represents a sample value at time t of the desired operation data of the j-th (j = 1, 2,..., J) type.

  The operation data acquisition unit 102 is, for example, a user's operation on a PC or TV keyboard key, a user's window selection operation, a user's operation on a button displayed on the GUI, a user's operation on a mouse button, a scroll bar Time-series data corresponding to the user's operation, such as the user's operation, that is, the time-series data representing the state of the user's operation (including the state where the user's operation is not performed) Acquired as desired operation data representing the operation desired by the user at that time.

  Note that what kind of data is used as the situation data and the desired operation data can be set in advance in the data processing apparatus of FIG. 1 or can be set by the user. it can.

  When requesting the user to set what kind of data is the situation data and the desired operation data, the situation data and the desired operation data such as the position of the mouse cursor and the user's operation on the key described above. A list of possible data may be displayed, and the user may select the status data and the desired operation data from the list of data.

  The prediction learning unit 103 learns the dynamics of the situation data from the situation data acquisition unit 101 and the desired operation data from the operation data acquisition unit 102 using the dynamics learning model stored in the dynamics learning model storage unit 104.

  That is, the predictive learning unit 103 uses parameters of the dynamics learning model stored in the dynamics learning model storage unit 104 using the situation data from the situation data acquisition unit 101 and the desired operation data from the operation data acquisition unit 102. The dynamics learning model to be updated is learned, whereby the dynamics learning model stored in the dynamics learning model storage unit 104 is updated to the situation data from the situation data acquisition unit 101 and the desired operation from the operation data acquisition unit 102. Acquire data dynamics.

  The dynamics learning model storage unit 104 stores a dynamics learning model that is a model that can acquire dynamics.

  Here, as a dynamics learning model, for example, NN (Neural Network) such as RNN (Recurrent Neural Network), FNN (Feed Forward Neural Network), and RNN-PB (Recurrent Neural Net with Parametric Bias), SVR (Support Vector Regression) and other models that can acquire dynamics can be adopted.

  As a model that can acquire a large number of dynamics, there is a dynamics storage network that includes a plurality of nodes and holds the dynamics in each of the plurality of nodes. The dynamics storage network will be described later.

The prediction unit 105 uses the situation data s ⁱ _t , s ⁱ _{t + 1} ,... Supplied from the situation data acquisition unit 101 based on the dynamics acquired by the dynamics learning model stored in the dynamics learning model storage unit 104. s ⁱ _{t + T} as input, predicted values a ′ ^j _t , a ′ ^j _{t + 1} ,..., a ′ ^j _{t + T} of the desired operation data, and, if necessary, prediction of situation data Values s ′ ^j _t , s ′ ^j _{t + 1} ,..., S ′ ^j _{t + T} are obtained and supplied to the operation data output unit 106.

When the prediction unit 105 obtains the predicted values s ′ ^j _t , s ′ ^j _{t + 1} ,..., S ′ ^j _{t + T} of the situation data, the situation data from the situation data acquisition unit 101 ( S ⁱ _t , s ⁱ _{t + 1} ,..., S ⁱ _{t + T} are also supplied to the operation data output unit 106.

The operation data output unit 106 uses the predicted values a ′ ^j _t , a ′ ^j _{t + 1} ,..., A ′ ^j _{t + T} of the desired operation data from the prediction unit 105 as one by one. The data is output to an interface (module) that receives operation data (data corresponding to an operation performed by the user (data indicating an operation)) such as a PC or TV as an electronic device constituting the unit.

The operation data output unit 106 uses the predicted values a ′ ^j _t , a ′ ^j _{t + 1} ,..., A ′ ^j _{t + T} of the desired operation data from the prediction unit 105 as operation data such as a PC or TV. When output to the accepting interface, the PC or TV performs processing according to the predicted values a ′ ^j _t , a ′ ^j _{t + 1} ,..., A ′ ^j _{t + T} of the desired operation data.

Note that the operation data output unit 106 receives the predicted values a ′ ^j _t , a ′ ^j _{t + 1} ,..., A ′ ^j _{t + T} of the desired operation data from the prediction unit 105 and the predicted values of the situation data. _{^{s' j t, s' j}} t + 1, ···, s' j t + T, and, true value s ⁱ _t of the situation _{^{data, s i t + 1, ···}} , is s ⁱ _{t + T} When supplied, the prediction error e ⁱ _t = | s ⁱ −s ′ ^j | of the predicted value of the situation data is obtained. Then, the operation data output unit 106, for example, each of the prediction errors e ^j _t , e ^j _{t + 1} ,..., E ^j _{t + T} or the total sum thereof is equal to or less than a predetermined threshold value determined in advance. Only when it is (less than), the predicted values a ′ ^j _t , a ′ ^j _{t + 1} ,..., A ′ ^j _{t + T} of the desired operation data from the prediction unit 105 are output.

In this case, when the prediction error is large, the operation data output unit 106 predicts the desired operation data from the prediction unit 105 a ′ ^j _t , a ′ ^j _{t + 1} ,..., A ′ ^j _{t + Since T} is not output, processing according to the predicted values a ′ ^j _t , a ′ ^j _{t + 1} ,..., A ′ ^j _{t + T} of the desired operation data is not performed on a PC, TV, or the like. That is, in PC, TV, etc., only when the prediction error is small, the process according to the predicted values a ′ ^j _t , a ′ ^j _{t + 1} ,..., A ′ ^j _{t + T} of the desired operation data is performed. Done.

  The control of the output of the desired operation data based on the prediction error as described above can be validated or invalidated according to a user operation, for example.

  Next, the situation data acquired by the situation data acquisition unit 101 of FIG. 1 and the desired operation data acquired by the operation data acquisition unit 102 will be further described with reference to FIG.

  FIG. 2A shows the locus of the position of the mouse cursor on the PC screen and the state where the icon at the upper left position of the PC screen is clicked.

  In FIG. 2A, the user moves the mouse cursor from the lower right to the upper left of the PC screen, and clicks on the icon at the upper left position of the PC screen.

  FIG. 2B shows status data acquired by the status data acquisition unit 101 when the operation of FIG. 2A is performed.

  In FIG. 2B, the horizontal axis represents time (time) t, and the vertical axis represents coordinates (x, y) in the xy coordinate system with the lower left point of the screen of FIG. 2A as the origin.

As shown in FIG. 2B, the situation data acquisition unit 101 uses coordinates (mouse_x _t , mouse_y _t ) representing the locus of the position of the mouse cursor as the i-th kind of situation data s ⁱ _t , s ⁱ _{t + 1} ,. · ·, s ⁱ acquires as _{t + T,} the coordinates (icon_x _t, icon_y _t) representative of the position of the icon in the upper left position of the screen of FIG. 2A, i 'th type of status data s ^i' _t , s ^{i ′} _{t + 1} ,..., s ^{i ′} _{t + T.}

  FIG. 2C shows desired operation data acquired by the operation data acquisition unit 102 when the operation of FIG. 2A is performed.

  In FIG. 2C, the horizontal axis represents time t, and the vertical axis represents mouse button off (off) and on (on).

As shown in FIG. 2C, the operation data acquisition unit 102 converts the time series data representing the ON / OFF operation state of the mouse button into the jth desired operation data a ^j _t , a ^j _{t + 1} ^,.・ ・ Acquire as a ^j _{t + T.}

  Next, in the data processing apparatus of FIG. 1, a dynamics learning model learning process (learning process) in which parameters of the dynamics learning model stored in the dynamics learning model storage unit 104 are updated using situation data and desired operation data. ) And a prediction process (prediction process) for obtaining a predicted value of the desired operation data (and a predicted value of the situation data) using the dynamics learning model that has acquired the dynamics through the learning process and using the situation data as input. Is called.

  FIG. 3 is a flowchart for explaining a learning process performed by the data processing apparatus of FIG.

  The learning process is started, for example, periodically or at irregular timing. In step S101, the situation data acquisition unit 101 acquires the situation data and supplies the situation data to the prediction learning unit 103, and the process is performed in step S102. Proceed to

  In step S102, the operation data acquisition unit 102 acquires desired operation data and supplies it to the prediction learning unit 103, and the process proceeds to step S103.

  In step S103, the predictive learning unit 103 reads the parameters of the dynamics learning model stored in the dynamics learning model storage unit 104, and the process proceeds to step S104.

  In step S <b> 104, the predictive learning unit 103 uses the dynamics learning model parameters read from the dynamics learning model storage unit 104 in step S <b> 103, the situation data from the situation data acquisition unit 101, and the desired operation from the operation data acquisition unit 102. After updating using the data, the process proceeds to step S105.

That is, for example, a three-layer NN having three layers of an input layer, a hidden layer (intermediate layer), and an output layer composed of an arbitrary number of units corresponding to neurons is adopted as a dynamics learning model. If so, the predictive learning unit 103 receives, as input data (vector) X _t at time t, status data s ⁱ _t at time t from the status data acquisition unit 101 and operation data as input data (vector) X _t. The desired operation data a ⁱ _t at time t from the acquisition unit 102 is input.

Thus, the units of the hidden layer, as the target input data X _t which is input to the input layer, the weighted addition using the connection weight for coupling the units to each other as the neuron (connection weight) conducted, further, the weighting An operation of a nonlinear function using the addition result as an argument is performed, and the operation result is output to the output layer unit.

From the unit of the output layer, as the output data corresponding to input data X _t, the predicted value X _{'t + 1} of the input data X _{t + 1} at the next time t + 1 of the input data X _t, namely, in this case, predicted value s status data at time _{t + 1} ^'i t + 1, and the predicted value a desired operating data at time _{t + 1'} ⁱ t + 1 is output.

Prediction learning unit 103 is, for example, BP accordance (Back-propagation) method, input of the output data to the data X _t, the following time t + 1 of the input data X _{t + 1} of the predicted value X of the input data X _t the _{'t + 1,} so the prediction error is smaller for the true value (time t + 1 of the input data X _{t + 1),} the connection weight as a parameter NN, a calculation for updating by a value corresponding to the prediction error Repeat until the NN parameters converge.

  Note that the update of the NN parameter is performed using the parameter read from the dynamics learning model storage unit 104 in step S103 as the initial value of the parameter.

As described above, from the time series data X _t at time t, learning to predict time series data X _{t + 1} at the next time t + 1 (prediction learning (prediction learning)) by performing, dynamics learning NN as a model can learn the time evolution law of time series data and acquire the dynamics of the time series data.

  In step S105, the predictive learning unit 103 writes (saves) the parameters of the dynamics learning model updated in step S104 in the form of overwriting in the dynamics learning model storage unit 104, and the process ends.

  FIG. 4 is a flowchart for explaining a prediction process performed by the data processing apparatus of FIG.

  The prediction process is started, for example, periodically or at irregular timing. In step S111, the situation data acquisition unit 101 acquires the situation data and supplies the situation data to the prediction unit 105, and the process proceeds to step S112. move on.

  In step S112, the prediction unit 105 reads the parameters of the dynamics learning model stored in the dynamics learning model storage unit 104, and the process proceeds to step S113.

  In step S113, the prediction unit 105 receives the situation data supplied from the situation data acquisition unit 101 based on the dynamics acquired by the dynamics learning model stored in the dynamics learning model storage unit 104, and predicts the desired operation data. Is supplied to the operation data output unit 106, and the process proceeds to step S114.

That is, for example, if the dynamics learning model is the three-layer NN described with reference to FIG. 3, the prediction unit 105 sends the status data s ^{i at} the time t from the status data acquisition unit 101 to the unit of the input layer. and _t, as desired operation data a ⁱ _t at time t, for example, the random number and a predetermined value such as, inputs as input data X _t at time t.

Thus, the units of the hidden layer, as the target input data X _t which is input to the input layer, is performed weighting addition using a link weight as the parameter dynamics learning model, further argument the result of the weighted addition And the result of the calculation is output to the output layer unit.

In the output layer unit, the calculation result of the hidden layer unit is input, and the same calculation as that of the hidden layer unit is performed. The calculation result is the input data X _t at the time t + 1 next to the input data X _{t. As} the predicted value X ′ _{t + 1 of +1} , that is, the predicted value s ′ ⁱ _{t + 1} of the situation data at time t + 1, and the predicted value a ′ ⁱ _{t + 1} of the desired operation data at time t + 1 Is output.

Then, the predicted value a ′ ⁱ _{t + 1} of the desired operation data at time t + 1 output from the output layer unit is supplied from the prediction unit 105 to the operation data output unit 106.

Incidentally, the unit of the input layer, after the situation data s ⁱ _t at time t from state data acquisition unit 101 is input, status data s of the next time t + 1 from state data acquisition unit 101 ⁱ _{t + 1} is input, and as the desired operation data a ⁱ _{t at} time t, for example, the predicted value a ′ ⁱ _{t + 1} of the desired operation data at time t + 1 output from the unit of the output layer is, for example, Used (input to the input layer unit).

In the foregoing paragraphs, the prediction unit 105, using a dynamics learning model, to the input of the input data X _t at time t, the input data X _{t + 1} at the next time t + 1 of the input data X _t of it was and obtaining a prediction value X _{'t + 1,} the prediction unit 105, a prediction value obtained by using the dynamics learning model, by repeating the further feedback as an input dynamics learning model, the time t In addition to the predicted value X ′ _{t + 1} of the input data X _{t + 1} of +1, predicted values X ′ _{t + 2} , X ′ _{t + 3} ,... Can also be obtained. The same applies to a prediction unit 165 (FIG. 9) described later.

In step S114, the operation data output unit 106 uses the predicted values a ′ ^j _t , a ′ ^j _{t + 1} ,..., A ′ ^j _{t + T} of the desired operation data from the prediction unit 105 as the data in FIG. The processing apparatus outputs the data to an interface that accepts operation data such as a PC or a TV as an electronic device that constitutes a part thereof, and the processing ends.

As a result, processing according to the predicted values a ′ ^j _t , a ′ ^j _{t + 1} ,..., A ′ ^j _{t + T} of the desired operation data is performed on a PC, TV, or the like.

  4 can be performed when the learning process of FIG. 3 is not performed, or can be performed in parallel with the learning process of FIG.

  As described above, the situation data obtaining unit 101 obtains situation data that is time-series data representing a situation, and the operation data obtaining unit 102 obtains desired operation data that is time-series data corresponding to an operation desired by the user. The prediction learning unit 103 learns the dynamics of the situation data and the desired operation data, and the prediction unit 105 obtains a predicted value of the desired operation data based on the dynamics, using the situation data as an input, and the operation data output unit In 106, since the predicted value of the desired operation data is output, for each user, the operation that the user is trying to perform, that is, the operation desired by the user can be performed in advance.

That is, for example, as shown in FIG. 2A, when the user moves the mouse cursor from the lower right to the upper left of the PC screen and clicks on the icon at the upper left position of the PC screen. In the situation data acquisition unit 101, as shown in FIG. 2B, coordinates (mouse_x _t , mouse_y _t ) representing the locus of the position of the mouse cursor are the i-th type of situation data s ⁱ _t , s ⁱ _{t. The} coordinates (icon_x _t , icon_y _t ) obtained as ₊₁ ,..., S ⁱ _{t + T} and representing the position of the icon at the upper left position of the screen in FIG. Data s ^{i ′} _t , s ^{i ′} _{t + 1} ,..., S ^{i ′} _{t + T} are acquired.

Furthermore, in the operation data acquisition unit 102, as shown in FIG. 2C, operation data representing an operation of turning on a mouse button that is turned off at a predetermined timing is j-th desired operation data a ^j _t , acquired as a ^j _{t + 1} ,..., a ^j _{t + T.}

In this case, the predictive learning unit 103 sets time series data of coordinates (mouse_x _t , mouse_y _t ) representing the locus of the mouse cursor from the lower right to the upper left of the PC screen, and the icon at the upper left position of the PC screen. Learning dynamics learning model using time-series data of coordinates (icon_x _t , icon_y _t ) representing the position of the mouse and time-series data representing an operation to turn on a mouse button that is turned off at a predetermined timing Is done.

As a result, the dynamics learning model uses time series data of coordinates (mouse_x _t , mouse_y _t ) representing the locus of the mouse cursor from the lower right to the upper left of the PC screen, and the icon at the upper left position of the PC screen. Time-series data of coordinates (icon_x _t , icon_y _t ) representing a position and dynamics of time-series data representing an operation of turning on a mouse button that is turned off at a predetermined timing are acquired.

Then, there the situation data acquiring unit 101, the coordinate (mouse_x _t, mouse_y _t) from the lower right represents the trajectory of the mouse cursor toward the direction of the upper left corner of the screen of the PC and the time-series data of the position of the upper left of the screen of the PC When time series data of coordinates (icon_x _t , icon_y _t ) representing the position of the icon is acquired and supplied to the prediction unit 105, the prediction unit 105 predicts the desired operation data having the dynamics acquired by the dynamics learning model. (Furthermore, the predicted value of the situation data), that is, the predicted value of the time-series data representing the operation of turning on the mouse button that is turned off at a predetermined timing, is obtained via the operation data output unit 106 Is output to the PC interface.

  Therefore, on a PC, when there is an icon at the upper left position of the PC screen, and the mouse cursor is in a situation that draws a trajectory from the lower right side of the PC screen toward the upper left direction. The predicted value of the desired operation data corresponding to the operation of clicking on the icon at the upper left position of the PC screen is obtained, and on this PC, the user clicks on the icon at the upper left position of the PC screen Before an operation is performed, processing according to the operation is performed in a form that preempts the operation.

  A dynamics learning model that can acquire dynamics, such as NN and SVR, has a so-called generalization ability (function). According to the dynamics learning model having the generalization ability, even when time-series data that does not match the time-series data used in the learning process is given, the given time-series data is converted into the learning process. When the time series data used in the above and the dynamics are similar, a prediction value with a certain degree of prediction accuracy is output.

  5 and 6 show other examples of situation data and desired operation data.

  That is, FIG. 5 shows the locus of the line of sight of the user who is viewing the article displayed on the web browser on the PC screen, and the operation of scrolling the web browser (display) after the locus is drawn. Shows the state.

In FIG. 5, the situation data acquisition unit 101 acquires coordinates representing the locus of the user's line of sight as i-th type situation data s ⁱ _t , s ⁱ _{t + 1} ,..., S ⁱ _{t + T.} At the same time, the operation data acquisition unit 102 acquires operation data corresponding to the scroll operation of the web browser as the jth desired operation data a ^j _t , a ^j _{t + 1} ,..., A ^j _{t + T} In this case, the prediction learning unit 103 learns the dynamics learning model using the time series data of the coordinates representing the locus of the user's line of sight and the time series data representing the scrolling operation of the web browser.

  As a result, the dynamics learning model acquires time-series data of coordinates representing the locus of the user's line of sight, and dynamics of time-series data representing an operation of scrolling the web browser.

  After that, when the situation data acquisition unit 101 acquires time series data of coordinates representing the locus of the user's line of sight as shown in FIG. 5 and supplies it to the prediction unit 105, the prediction unit 105 uses the dynamics learning model. The predicted value of the desired operation data having the dynamics acquired by the user (and the predicted value of the situation data), that is, the predicted value of the time-series data representing the scrolling operation of the web browser is obtained. To the PC interface.

  Therefore, in the PC, when the user's line of sight on the web browser is in a situation where a locus as shown in FIG. 5 is drawn, a predicted value of time series data representing an operation of scrolling the web browser is obtained, Thus, on the PC, before the user performs an operation for scrolling the web browser, processing in accordance with the operation, that is, scrolling of the web browser is performed in advance.

  FIG. 6 shows a TV screen (display screen).

  That is, in FIG. 6, the TV screen is divided into a main screen and a sub screen on the left and right, and the image of the program selected by the user is displayed on the main screen, and four so-called back programs are displayed on the sub screen. Four reduced images obtained by reducing the respective images are displayed in a vertically arranged form.

  Furthermore, in FIG. 6, the locus of the user's line of sight that reciprocates between the image of the program on the main screen and the reduced image of the second program from the top of the sub screen, and after the locus is drawn, An operation for selecting the program on which the second reduced image is displayed from the top is performed.

In FIG. 6, the situation data acquisition unit 101 acquires the coordinates representing the user's line of sight as the i-th kind of situation data s ⁱ _t , s ⁱ _{t + 1} ,..., S ⁱ _{t + T.} At the same time, the operation data acquisition unit 102 obtains the operation data corresponding to the operation for selecting the program on which the second reduced image is displayed from the top of the sub screen, as the jth desired operation data a ^j _t , a ^j _{t + 1.} ,..., ^Aj _{t + T} , the prediction learning unit 103 displays the time-series data of coordinates representing the user's line of sight and the second reduced image from the top of the sub-screen. Learning of a dynamics learning model is performed using time series data representing an operation of selecting a program.

  As a result, the dynamics learning model has coordinates of the user's line of sight (the locus of the user's line of sight that reciprocates between the program image on the main screen and the reduced image of the second program from the top of the sub screen). Time-series data and time-series data representing an operation for selecting the program on which the second reduced image is displayed from the top of the sub screen (selection before and after the operation for selecting the program (selection operation)) Acquire dynamics of time-series data indicating that no operation has been performed.

  After that, when the situation data acquisition unit 101 acquires time series data of coordinates representing the locus of the user's line of sight as shown in FIG. 6 and supplies the time series data to the prediction unit 105, the prediction unit 105 uses the dynamics learning model. Is the predicted value of the desired operation data having the dynamics acquired by (and also the predicted value of the situation data), that is, the time-series data representing the operation of selecting the program on which the second reduced image is displayed from the top of the sub-screen. A predicted value is obtained and output to the TV interface via the operation data output unit 106.

  Therefore, in the TV, when the user's line of sight on the screen is in a state of drawing a locus as shown in FIG. 6, an operation for selecting the program on which the second reduced image is displayed from the top of the sub screen is performed. The predicted value of the time-series data to be expressed is obtained, so that, on the TV, before the user performs the operation of selecting the program on which the second reduced image is displayed from the top of the sub screen, the operation is preempted. Thus, processing according to the operation, that is, displaying the image of the program on which the second reduced image is displayed from the top of the sub screen is performed on the main screen.

In addition, in the prediction process of FIG. 4, the prediction unit 105 (FIG. 1), as described above, the situation data s ⁱ _t , s ⁱ _{t + 1} ,..., S ⁱ supplied from the situation data acquisition unit 101. _{Using t + T} as an input, the predicted value a ′ ^j _t , a ′ ^j _{t + 1} ,..., a ′ ^j _{t + T} of the desired operation data is obtained, and the predicted value s ′ ^j _t , s of the situation data ' ^j _{t + 1} , ..., s' ^j _{t + T} is obtained, and the situation data (true value) s ⁱ _t , s ⁱ _{t + 1} , ..., s supplied from the situation data acquisition unit 101 It can be supplied to the operation data output unit 106 together with ⁱ _{t + T.}

In this case, the operation data output unit 106 includes the predicted values s ′ ^j _t , s ′ ^j _{t + 1} ,..., S ′ ^j _{t + T} of the situation data from the prediction unit 105 and the true value of the situation data. _{^{s i t, s i t +}} 1, ···, s i t + T using a prediction of the predicted value of the situation data error _{^{e i t = | s i -s}} ' j | look, the prediction error e ^{j _{t, e j t + 1,}} ···, each e ^j _{t + T,} or, all of the sum, only when it is below a predetermined threshold value that is predetermined, the prediction of the desired operating data from the prediction unit 105 The values a ′ ^j _t , a ′ ^j _{t + 1} ,..., A ′ ^j _{t + T} are output.

In this case, when the predicted values s ′ ^j _t , s ′ ^j _{t + 1} ,..., S ′ ^j _{t + T of} the situation data are not certain, eventually, the predicted values a ′ ^j _t , When a ′ ^j _{t + 1} ,..., a ′ ^j _{t + T} are not certain, the predicted value a ′ ^j _t , a ′ ^j _{t + 1} ,. It is possible to prevent the processing according to a ′ ^j _{t + T from} being performed, that is, the processing not intended by the user.

  Next, FIG. 7 is a block diagram showing a configuration example of another embodiment of the data processing apparatus to which the present invention is applied.

  The data processing device in FIG. 7 constitutes a part of a game device (or game software), for example.

  Here, it is assumed that the game device that is a part of the data processing device of FIG. 7 is a game device for games such as soccer and baseball, in which a user operates a plurality of characters.

7, the data processing apparatus includes a teaching character selection unit 151 and N character motion auxiliary modules 152 ₁ , 152 ₂ ,..., The same number as a plurality of N characters operated by the user. 152 _N.

The teaching character selection unit 151 selects, for example, a character for learning a motion from N characters in response to a user operation, and a character motion assist module 152 _n that is responsible for assisting the motion of the character. Control to perform the learning process.

As described above, the character motion assist modules 152 ₁ to 152 _N are provided in the same number N as the N characters operated by the user. The character motion assist module 152 _n is responsible for assisting the motion of the nth character among the N characters.

That is, the character motion assist module 152 _n learns the motion of the nth character and assists the motion of the nth character according to the learning result.

  FIG. 8 shows a game screen (game screen) displayed by a game device that constitutes a part of the data processing device of FIG.

  That is, FIG. 8 shows a game screen of a soccer game.

  In FIG. 8, four teammate characters FC # 1, FC # 2, FC # 3, and FC # 4, which are part of the user's team, are the characters of the soccer player, and the enemy controlled by the game device. Three enemy characters RC # 1, RC # 2, and RC # 3, which are part of the team character, are displayed.

  Further, in FIG. 8, a soccer ball is displayed. During the game, for example, as a general rule, the ally character FC # n closest to the ball moves according to the user's operation, and other ally The character FC # n ′ is operated by the game device.

  In FIG. 8, among the teammate characters FC # 1 to FC # 4, the teammate character FC # 1 is closest to the ball. Therefore, the teammate character FC # 1 moves according to the user's operation, and other teammate characters FC # 2 to FC # 4 operate according to the control of the game device.

  It is programmed in advance in the game software how the game device operates other teammate characters FC # n '. Therefore, for example, in FIG. 8, while the user is operating the teammate character FC # 1, the other teammate characters FC # 2 to FC # 4 want to move as indicated by arrows. However, if such programming is not performed in the game software, the other ally characters FC # 2 to FC # 4 do not move.

Therefore, the character motion assist module 152 _n learns the motion of the n th character, and assists the motion of the n th character according to the learning result, so that the motion desired by the user ( If the user is operating, an operation corresponding to an operation that will be performed) is performed.

That is, FIG. 9 shows a configuration example of the character motion assist module 152 _n of FIG.

In FIG. 9, the character motion assist module 152 _n includes a situation data acquisition unit 161, an operation data acquisition unit 162, a prediction learning unit 163, a dynamics learning model storage unit 164, a prediction unit 165, and an operation data output unit 166. .

The situation data acquisition unit 161 acquires situation data s ⁱ _t , s ⁱ _{t + 1} ,..., S ⁱ _{t + T} that are time-series data representing the situation, as in the situation data acquisition unit 101 of FIG. And supplied to the prediction learning unit 163 and the prediction unit 165.

  However, the situation data acquisition unit 161, for example, indicates the position of the teammate characters FC # 1 to FC # 4, the position of the ball, the position of the goal, the position of the enemy characters RC # 1 to RC # 3, etc. Acquire situation data representing various situations.

Similar to the operation data acquisition unit 101 in FIG. 1, the operation data acquisition unit 162 is desired operation data a ^j _t , a ^j _{t + 1} ,..., A that is time-series data corresponding to an operation desired by the user. ^j _{t + T} is acquired and supplied to the prediction learning unit 163.

However, the operation data acquisition unit 162, for example, the operation and the user moves the n-th character is the character operation auxiliary module 152 _n in FIG charge (ally character), its n-th character, shoot, pass, Desired operation data representing an operation for causing the nth character to perform a motion desired by the user, such as a user operation for dribbling, is acquired.

  Similar to the prediction learning unit 103 in FIG. 1, the prediction learning unit 163 stores the dynamics of the situation data from the situation data acquisition unit 161 and the desired operation data from the operation data acquisition unit 162 in the dynamics learning model storage unit 164. Learning with a memorized dynamics learning model.

  The dynamics learning model storage unit 164 stores a dynamics learning model in the same manner as the dynamics learning model storage unit 104 in FIG.

Like the prediction unit 105 in FIG. 1, the prediction unit 165 uses the situation data s ⁱ _t , supplied from the situation data acquisition unit 161 based on the dynamics acquired by the dynamics learning model stored in the dynamics learning model storage unit 164. s ⁱ _{t + 1} ,..., s ⁱ _{t + T} as input, the predicted values a ′ ^j _t , a ′ ^j _{t + 1} ,..., a ′ ^j _{t + T} of the desired operation data, and if necessary, the predicted value s' ^j _t, s' of the status data obtains a _{^{j t + 1, ···, s}} ' j t + T, and supplies the operation data output unit 166.

In addition, when the prediction unit 165 obtains the predicted values s ′ ^j _t , s ′ ^j _{t + 1} ,..., S ′ ^j _{t + T} of the situation data, the situation data from the situation data acquisition unit 161 ( S ⁱ _t , s ⁱ _{t + 1} ,..., S ⁱ _{t + T} are also supplied to the operation data output unit 166.

The operation data output unit 166 is similar to the operation data output unit 106 of FIG. 1, and predicted values a ′ ^j _t , a ′ ^j _{t + 1} ,..., A ′ ^j _t of the desired operation data from the prediction unit 165. _{+ T} is output to an interface (module) that receives operation data of a game device that is a part of the data processing device of FIG.

An interface in which the operation data output unit 166 receives the operation data of the game device using the predicted values a ′ ^j _t , a ′ ^j _{t + 1} ,..., A ′ ^j _{t + T} of the desired operation data from the prediction unit 165 In the game device, the game device performs a process of moving the n-th character according to the predicted values a ′ ^j _t , a ′ ^j _{t + 1} ,..., A ′ ^j _{t + T} of the desired operation data. .

In operation data output unit 166, similarly to the operation data output unit 106 of FIG. 1, the prediction error e ⁱ _t = the predicted value of the status data | s ⁱ -s' ^j | look, the prediction error e ^j _t, Only when the sum of all or all of e ^j _{t + 1} ,..., e ^j _{t + T} is less than (or less than) a predetermined threshold value determined in advance, the desired operation data from the prediction unit 165 Predicted values a ′ ^j _t , a ′ ^j _{t + 1} ,..., A ′ ^j _{t + T} can be output.

  Next, the situation data acquired by the situation data acquisition unit 161 in FIG. 9 and the desired operation data acquired by the operation data acquisition unit 162 will be further described with reference to FIG.

  FIG. 10A shows a game screen in the learning mode.

  That is, in the game device in which the data processing device of FIG. 7 constitutes a part, the normal mode for playing a soccer game with the enemy team operated by the game device (or another user) as the operation mode, There is a learning mode for learning the movement of the character.

  FIG. 10A shows a game screen when the operation mode is the learning mode. Note that the operation mode is switched in accordance with, for example, a user operation.

  FIG. 10A shows the same game screen as in FIG.

  Therefore, in the game screen of FIG. 10A, four friendly characters FC # 1 to FC # 4 that are part of the characters of the user's team and three characters that are part of the enemy team operated by the game device are displayed. The enemy characters RC # 1 to RC # 3 exist, and among the teammate characters FC # 1 to FC # 4, the teammate character FC # 1 is located closest to the ball. ing.

  In the situation of FIG. 10A, when the user wants to move each of the teammate characters FC # 2 to FC # 4 other than the teammate character FC # 1 closest to the ball as indicated by the arrows, for example, the teammate character FC # 2 is selected as an action learning target character, and an operation of moving the friend character FC # 2 as indicated by an arrow is performed.

In this case, teaching character selection section 151 (FIG. 7) is responsible for assisting the operation of the teammate character FC # 2, for example, the second character operation auxiliary module 152 ₂ is controlled to perform the learning process. Character movement auxiliary module 152 ₂ under the control of the teaching character selection section 151, the status data obtaining unit 161 obtains the status data, the operation data acquisition unit 162, the desired operating data (here, the teammate character FC # 2 10A is acquired, and learning processing is performed for learning the dynamics of the situation data and desired operation data.

FIG. 10B shows a character motion assistance that assists the movement of the teammate character FC # 2 when the user performs an operation of moving the teammate character FC # 2 as indicated by an arrow on the game screen of FIG. 10A. It shows a situation data status data obtaining unit 161 of the module 152 ₂ obtains.

  In FIG. 10B, the horizontal axis represents time t, and the vertical axis represents coordinates (x, y) in the xy coordinate system with the lower left point of the game screen of FIG. 10A as the origin.

For example, as shown in FIG. 10B, the situation data acquisition unit 161 uses coordinates (ball_x _t , ball_y _t ) representing the trajectory of the position of the ball as i-th type situation data s ⁱ _t , s ⁱ _{t + 1} , ..., obtained as s ⁱ _{t + T.}

FIG. 10C shows a character motion assistance that assists the movement of the teammate character FC # 2 when the user performs an operation of moving the teammate character FC # 2 as indicated by an arrow on the game screen of FIG. 10A. It indicates the desired operation data operation data acquisition unit 162 of the module 152 ₂ obtains.

  In FIG. 10C, the horizontal axis represents time t, and the vertical axis represents the xy coordinate system having the origin at the lower left point of the game screen of FIG. The coordinates (x, y) are represented.

As shown in FIG. 10C, the operation data acquisition unit 162 represents coordinates (p_move_x _t , p_move_y _t ) representing the locus of the position of the teammate character FC # 2 that is a character to be learned for movement and moves according to the user's operation. Obtained as j-th desired operation data a ^j _t , a ^j _{t + 1} ,..., a ^j _{t + T.}

Hereinafter, similarly, the user performs an operation of selecting each of the teammate characters FC # 3 and FC # 4 as a character to be learned for movement and moving it as indicated by an arrow. Thus, for example, the third character motion assist module 152 ₃ and the fourth character motion assist module 152 ₄ are responsible for assisting the motion of the teammate characters FC # 3 and FC # 4. _{As in} the case of ₂ , situation data and desired operation data are acquired.

  Next, the operation of the data processing apparatus in FIG. 7 will be described.

In the data processing device of FIG. 7, as in the data processing device of FIG. 1, in the character motion assist module 152 _n (FIG. 9), the parameters of the dynamics learning model stored in the dynamics learning model storage unit 164 are used as the situation data and The dynamics learning model learning process (learning process), which is updated using the desired operation data, and the dynamics learning model that has acquired the dynamics through the learning process, using the situation data as input, the predicted value of the desired operation data (and more , A prediction process (prediction process) for obtaining a predicted value of the situation data is performed.

  FIG. 11 is a flowchart illustrating the learning process performed by the data processing apparatus of FIG.

  The learning process is performed, for example, when the operation mode is set to the learning mode.

In step S151, the teaching character selection unit 151 waits for the user to perform an operation of selecting a character for learning a motion, and learns the character motion assist module 152 _n responsible for assisting the motion of the character. It selects as a learning object module, and a process progresses to step S152.

In step S152, in the character motion assistance module 152 _n (FIG. 9) selected as the learning target module, the situation data acquisition unit 161 acquires situation data and supplies the situation data to the prediction learning unit 163, and the processing is performed in step S153. Proceed to

  In step S153, the operation data acquisition unit 162 acquires the desired operation data, supplies it to the prediction learning unit 163, and the process proceeds to step S154.

  In step S154, the prediction learning unit 163 reads the parameters of the dynamics learning model stored in the dynamics learning model storage unit 164, and the process proceeds to step S155.

  In step S155, the prediction learning unit 163 uses the dynamics learning model parameters read from the dynamics learning model storage unit 164 in step S154, the situation data from the situation data acquisition unit 161, and the desired operation from the operation data acquisition unit 162. After updating using the data, the process proceeds to step S156.

  In step S156, the prediction learning unit 163 writes the parameters of the dynamics learning model updated in step S155 in the form of overwriting in the dynamics learning model storage unit 164, and the process ends.

  FIG. 12 is a flowchart for explaining a prediction process performed by the data processing apparatus of FIG.

For example, when the motion mode is set to the normal mode, the prediction process includes N character motion assist modules 152 ₁ to 152 _N (however, the character motion assist module 152 _n ′ in charge of the character operated by the user). Can be excluded).

In the character motion assist module 152 _n , in step S161, the situation data acquisition unit 161 acquires situation data and supplies the situation data to the prediction unit 165, and the process proceeds to step S162.

  In step S162, the prediction unit 165 reads the parameters of the dynamics learning model stored in the dynamics learning model storage unit 164, and the process proceeds to step S163.

  In step S163, the prediction unit 165 receives the situation data supplied from the situation data acquisition unit 161 based on the dynamics acquired by the dynamics learning model stored in the dynamics learning model storage unit 164, and predicts the desired operation data. Is obtained and supplied to the operation data output unit 166, and the process proceeds to step S164.

  In step S164, the operation data output unit 166 outputs the predicted value of the desired operation data from the prediction unit 165 to the interface that receives the operation data of the game device of which the data processing device of FIG. The process ends.

Thereby, in the game device, a process of causing the character in charge of the character motion assist module 152 _n to move according to the predicted value of the desired operation data is performed.

As described above, in the character motion assist module 152 _n , the situation data acquisition unit 161 acquires situation data that is time-series data representing the situation, and the operation data acquisition unit 162 corresponds to an operation desired by the user. Desired operation data that is time-series data is acquired, the prediction learning unit 163 learns the dynamics of the situation data and the desired operation data, and the prediction unit 165 predicts the desired operation data using the situation data as an input based on the dynamics. Since the value is obtained and the predicted value of the desired operation data is output in the operation data output unit 166, an operation desired by the user, that is, an operation desired by the user can be performed in advance for each user.

  That is, for example, as shown in FIGS. 8 and 10, while the user is operating the teammate character FC # 1, the other teammate characters FC # 2 to FC # 4 are indicated by arrows. When you want to perform an operation to move to, you can move that way.

As described above, the operation data output unit 166 obtains the prediction error e ⁱ _t = | s ⁱ −s ′ ^j | of the predicted value of the situation data, and similarly to the operation data output unit 106 of FIG. prediction error _{^{e j t, e j t +}} 1, ···, e j t + according to _T, the predicted value a ^'j _t, a' desired operation data _{^{j t + 1, ···, a}} 'j By controlling the output of _{t + T} , it is possible to prevent a process unintended by the user, that is, a character from performing an unintended action.

  Next, as the dynamics learning model, as described above, a dynamics storage network that can hold a large number of dynamics can be employed.

  Therefore, in the following, a dynamics storage network will be described taking an example of application to an autonomous agent that acts autonomously such as an autonomous robot.

  Autonomous agents such as autonomous robots act autonomously by deciding how to behave based on various sensor signals, that is, determining the action to be taken and generating a motor signal according to the action. To do.

  Here, the sensor signal is, for example, an image signal output when the camera performs imaging as sensing, an audio signal output when the microphone (microphone) collects sound as sensing, or the like. The motor signal is, for example, a signal given to a motor that drives an arm or leg of an autonomous agent, a signal necessary for speech synthesis, etc. given to a speech synthesizer.

  When an autonomous agent decides an action to be taken, based on the sensor signal, the surrounding state (for example, the position where an object is present), the state of the autonomous agent (for example, the state of an arm or a leg, etc.) Recognize the situation. Recognizing this situation is hereinafter also referred to as recognition as appropriate.

  Further, the autonomous agent determines an action (motion) to be taken based on the recognition (recognition) result, and generates a motor signal corresponding to the action. When this motor signal is given to the motor that drives the arms, legs, etc. of the autonomous agent, the autonomous agent takes action to move the arms, legs, etc.

  Here, the generation of a motor signal corresponding to an action to be taken as appropriate is hereinafter simply referred to as an action.

  In addition, hereinafter, the situation is recognized as appropriate, the action to be taken is determined based on the recognition result, and the motor is generated according to the action. To act is also called cognitive behavior. A model that models cognitive behavior is called a cognitive behavior model.

  The cognitive behavior of autonomous agents can be described as dynamical systems defined by the law of time evolution, and it is known that various behaviors can be realized by specific attractor dynamics of the dynamic system. Yes. For example, the walking motion of a biped robot imitating a person can be described as limit cycle dynamics, which is characterized by the movement state of the system from various initial states to a specific periodic trajectory. it can.

  In addition, as an autonomous agent, for example, a leaching movement in which an arm robot extends a hand against an object, a fixed point dynamics (fixed- point dynamics). Furthermore, it is said that all the movements can be realized by a combination of a discrete movement that can be realized by the fixed point dynamics and a cyclic movement that can be realized by the limit cycle dynamics.

  Therefore, a dynamics learning model for learning dynamics can be used as a cognitive behavior model.

  One of the dynamics learning models, for example, RNN has a context unit connected to the network by a regression loop, and can hold any internal state in theory, so it can theoretically approximate any dynamic system. is there.

  However, it may be difficult to acquire (learn) a large number of dynamics with one RNN because of the convergence of learning.

  On the other hand, according to the dynamics storage network, a large number of dynamics can be easily acquired.

  FIG. 13 shows an example of the configuration of a data processing apparatus that learns to acquire dynamics of time-series data using a dynamics storage network, and recognizes and generates time-series data using the learning results.

  In the data processing apparatus of FIG. 13, an observation signal that can be observed is input to the signal input unit 11. The observation signal is, for example, a sound or image signal, the brightness of an LED (Light Emitting Diode), the rotation angle or the rotation angular velocity of the motor, and the data processing device of FIG. 13 is used for, for example, the recognition behavior of an autonomous agent. If so, a signal that can be input to and output from the autonomous agent can be an observation signal.

  Here, when the dynamics storage network is adopted as the dynamics learning model, the situation data and the desired operation data correspond to the observation signal.

  The signal input unit 11 outputs an electrical signal corresponding to the observed signal. Specifically, for example, the signal input unit 11 corresponds to a microphone as a sensor when the observation signal is a sound signal, and corresponds to a camera as a sensor when the observation signal is an image signal. Further, a measuring device for the rotation angle and rotation speed of the motor also corresponds to the signal input unit 11.

  Here, hereinafter, a signal input to the signal input unit 11 and a signal output from the signal input unit 11 are also referred to as observation signals as appropriate.

  Note that the observation signal may be a stationary signal that is stationary in time, or an unsteady signal that changes in time (not stationary).

  In the following, for example, a sensor motor signal is used as an observation signal. The sensor motor signal is, for example, a constant of a sensor signal output by a camera, microphone, or other sensor of an autonomous robot (not shown) and a motor signal given to a motor that drives the arm or leg of the autonomous robot. This is a vector time series obtained by sampling at a time interval of the same and having sample values at the same sample point (time) as components.

  The signal input unit 11 sequentially outputs the observation signals, which are time series data, divided into appropriate lengths. That is, the signal input unit 11 extracts, for example, 100 samples (points) from the sensor motor signal as the observation signal while shifting one sample at a time, and the time-series data of the 100 samples is extracted to the feature amount extraction unit 12. Supply.

  Note that the sampling time interval of the sensor motor signal and the number of samples (number of samples) extracted from the sensor motor signal by the signal input unit 11 are appropriately adjusted according to the sensor motor signal used as the observation signal.

  The feature amount extraction unit 12 extracts a feature amount from the observation signal supplied from the signal input unit 11, and supplies a time series of the feature amount to the learning unit 13, the recognition unit 14, and the generation unit 15.

  That is, when the observation signal is, for example, a voice signal, the feature amount extraction unit 12 performs frequency analysis and other acoustic processing every predetermined time of the voice signal, and is widely used for voice recognition and the like. For example, a voice feature amount such as a mel cepstrum is extracted. Then, the feature amount extraction unit 12 outputs the feature amount extracted from the observation signal in time series, and thereby the feature amount extraction unit 12 provides the feature to the learning unit 13, the recognition unit 14, and the generation unit 15. Quantity time-series data is supplied.

  The learning unit 13 performs a learning process for learning dynamics based on the time series data from the feature amount extraction unit 12, and the recognition unit 14 and the generation unit 15 learn based on the time series data from the feature amount extraction unit 12. Recognition processing that recognizes time-series data, generation processing that generates time-series data, recognition processing that recognizes time-series data and generates time-series data according to the recognition result I do.

  That is, the learning unit 13 performs a learning process for self-organizingly updating each dynamics of the dynamics storage network stored in the network storage unit 16 to be described later, based on the time series data from the feature amount extraction unit 12.

  Here, in the learning process, the parameters of the dynamics storage network are updated. The parameter update is also called learning.

  Although the details of the learning process by the learning unit 13 will be described later, in the learning process, basically, time-series data to which no label (correct answer label) is assigned is repeatedly given to the dynamics storage network (supply). Then, unsupervised learning is performed in which the dynamics storage network acquires the characteristic dynamics in the time-series data in a self-organizing manner. As a result, the dynamics storage network stores representative dynamics of the time-series data given thereto.

  Here, the dynamics storage network holds the dynamics by, for example, an RNN that is one of the dynamic system approximation models, as will be described later. For example, in response to an input of data at a certain time t, an input of data at a certain time t input to an RNN that outputs data at the next time t + 1 (an input unit of an input layer described later) When the data at the time t + 1 output by the RNN with respect to the data at the time t is referred to as output data, the feature amount extraction is performed for the learning unit 13, the recognition unit 14, and the generation unit 15. The time series data supplied from the unit 12 is input data.

  The recognition unit 14 performs a recognition process using the input data, that is, the time-series data supplied from the feature amount extraction unit 12 as a recognition target.

  That is, the recognizing unit 14 determines the dynamics most suitable for the time-series data supplied from the feature amount extracting unit 12 among the dynamics stored in the dynamics storage network of the network storage unit 16 and represents the dynamics. Information is output as a recognition result of time-series data as input data.

  In addition to the time-series data supplied from the feature amount extraction unit 12, the generation unit 15 is also supplied with a control signal. The generation unit 15 determines the dynamics used for generating the time series data from the dynamics stored in the dynamics storage network of the network storage unit 16 according to the control signal supplied thereto, and the time series having the dynamics A generation process for generating data using the time-series data supplied from the feature amount extraction unit 12 as necessary is performed.

  The time-series data obtained by the generation unit 15 performing the generation process is output after being subjected to necessary processes.

  The network storage unit 16 stores a dynamics storage network.

  The dynamics storage network holds dynamics in one node and is composed of a plurality of nodes.

  Here, the dynamics represents a dynamic system that changes with time, and can be expressed by a specific function, for example. In the dynamics storage network, the temporal change characteristics of time-series data are stored as dynamics.

  In this embodiment, the dynamics modeled by a dynamical approximate model having an internal state quantity, for example, is held in a node of the dynamics storage network. In this case, the dynamics storage network is a network having a dynamic system approximation model having an internal state quantity as a node (a network configured by nodes holding (storing) a dynamic system approximation model having an internal state quantity).

  Here, models with internal state quantities (dynamic system approximation) are, for example, inputs and outputs that can be observed from the outside when considering a model that outputs when there is an input. Apart from this, it is a model having an internal state quantity that cannot be observed from the outside and that represents the internal state of the model. In the model having the internal state quantity, the output is obtained using the internal state quantity in addition to the input. Therefore, even if the same input exists, a different output is obtained if the internal state quantity is different.

  The internal state storage unit 17 stores the internal state amount of the dynamics storage network stored in the network storage unit 16. The internal state quantity stored in the internal state storage unit 17 is appropriately updated when the time series data is recognized in the recognition generation process, and is used as necessary when generating the time series data. By this recognition generation process, the cognitive behavior of the autonomous agent can be realized.

  Next, FIG. 14 schematically shows an example of the dynamics storage network stored in the network storage unit 16 of FIG.

  The dynamics storage network is composed of a plurality of nodes and links.

  The node holds (stores) dynamics.

  A link provides a coupling relationship between nodes.

In FIG. 14, the dynamics storage network has nine nodes N ₁ to N ₉ , and each node N _i (i = 1, 2,..., 9) has nine nodes N ₁ to N 9. Links are provided between nodes adjacent in the vertical direction and the horizontal direction so that ₉ is arranged in a two-dimensional lattice pattern.

That is, in FIG. 14, a two-dimensional arrangement structure is given to _nine nodes N ₁ to N ₉ by links.

Here, in the dynamics storage network can, depending on the arrangement of the nodes N _i, defines the coordinate system representing the position of the node N _i. That is, for example, as shown in FIG. 14, the node N _i of a two-dimensional arrangement structure, defines a two-dimensional coordinate system, the coordinates on the two-dimensional coordinate system, is possible to express the node position N _i it can.

For example, in the dynamics storage network of FIG. 14, the two-dimensional configuration is such that the position of the lower left node N ₇ is the origin (0,0), the left to right direction is the x axis, and the bottom to upper direction is the y axis. If the coordinate system is defined and the link length is 0.5, for example, in the dynamics storage network of FIG. 14, the coordinates of the position of the upper right node N ₃ are (1, 1).

Also, as a measure representing the degree of similarity (closeness) between the dynamics held by any two nodes N _i and N _j constituting the dynamics storage network, the distance between the nodes N _i and N _j is Introduce.

Now, as the distance between the node N _i and N _j, When adopting the Euclidean distance between the node N _i and N _j, for example, the lower left node N _7, the upper right of the node N ₃ The distance between them is √ ((0-1) ² + (0-1) ² ) = √2.

Figure 15 shows schematically an example of the configuration of the network storage unit 16 to the node N _i of stored dynamics storage network of Figure 13.

The node _Ni has, for example, a dynamic system approximate model that has an internal state quantity and can represent a dynamic system.

  In FIG. 15, for example, RNN is adopted as a dynamic system approximation model having an internal state quantity.

  In FIG. 15, the RNN as a dynamical system approximation model is a three-layer NN having a regression loop from its output layer to its input layer, and the internal state quantity is retained by the regression loop. The

  That is, in FIG. 15, the RNN as a dynamic system approximation model is composed of three layers: an input layer, a hidden layer (intermediate layer), and an output layer. Each of the input layer, the hidden layer, and the output layer is configured by an arbitrary number of units corresponding to neurons.

  In FIG. 15, the input layer has an input unit and a context unit.

Time series data as input data (vector) X _{t at} time t is input to the input unit.

  For example, data output by some units in the output layer is fed back to the context unit as a context that is an internal state quantity. That is, in the RNN of FIG. 15, the context unit and a part of the output layer are connected by the regression loop, and the data output by the unit of the output layer is stored in the context unit. Is input as a context.

Here, when the input data X _t at time t is input to the input unit, the context C _t at time t is input to the context unit, the input data X _t-1 of one time before the time t-1 On the other hand, it is data output by some units in the output layer. Therefore, the data output by some units in the output layer with respect to the input of the input data X _t at time t becomes the context C _{t + 1 at} the next time t + 1.

The hidden layer unit performs weighted addition using connection weights (connection weights) that connect the units as neurons for the input data X _t and context C _t input to the input layer, and the weighted addition. A non-linear function is calculated using the result of the above as an argument, and the calculation result is output to the unit of the output layer.

As described above, data serving as the context C _{t + 1} at the next time t + 1 is output from some units in the output layer and fed back to the context unit in the input layer. Further, from the remaining units of the output layer, for example, as output data to the input data X _t, the predicted value X _{'t + 1} of the input data X _{t + 1} at the next time t + 1 of the input data X _t Is output.

In the RNN as described above, time-series data as input data is used as learning data for learning of the RNN, and from the time-series data X _t at the time t, the time-series data X t + 1 at the next time _{t + 1} By learning to predict (prediction learning), the time evolution law of learning data (time-series data) can be learned.

  Here, for example, a BPTT (Back-Propagation Through Time) method can be adopted as a learning method for obtaining the parameters of the dynamical approximate model having an internal state quantity such as RNN. For the BPTT method, see DE Rumelhart, GE Hinton & RE Williams, 1986 "Learning internal representations by error propagation", In DE Rumelhart & J. McClelland, "Parallel distributed processing, pp. 318-364, Cambridge, MA: MIT Press, RJ Williams and D. Zipser, “A learning algorithm for continuously running fully recurrent neural networks”, Neural Computation, 1: 270-280, 1989, and the like.

  The learning unit 13 updates the connection weight, which is a parameter of the RNN, so that the dynamics retained by the RNN as the dynamical system approximation model is affected by the learning data that is time-series data used for learning of the dynamics storage network. Learn RNN.

  The learning unit 13 has an adjustment function for increasing or decreasing the degree to which the dynamics held by the dynamical approximate model is affected by the learning data.

  That is, in learning of the dynamics storage network, every time learning data is input, the parameters of the dynamic system approximation model of the nodes constituting the dynamics storage network are updated little by little. At the time of updating this parameter, the learning unit 13 determines, for each node, a learning weight that represents the degree to which the dynamics held by the node are updated, that is, the degree to which the learning data affects the dynamics held by the node.

  The learning unit 13 updates the dynamics held by the node in a self-organized manner so as to be close to the dynamics of the learning data according to the learning weight.

  That is, the learning unit 13 adjusts the degree to which the dynamics held by the RNN is affected by the learning data when the dynamic system approximation model of the node is, for example, an RNN, Update the RNN parameters using the BPTT method.

  In the learning unit 13, the adjustment of the degree to which the dynamics held by the RNN are affected by the learning data is adjusted by, for example, setting the number of iterations for calculating the parameter when updating the RNN parameter (binding weight) by the BPTT method as the learning weight. For example, the prediction error that affects the degree of parameter update is corrected according to the learning weight.

That is, updates the parameters of the RNN by the BPTT method, for example, input data X as the output data for _t, the input data X _t of the input data of the next time _{t + 1} X t + 1 of the predicted value X _{'t + The} calculation to update the coupling weight as a parameter of RNN by a value corresponding to the prediction error so that the prediction error for the true value of ₁ (input data X _{t + 1 at} time t + 1) is small is Repeated until parameters converge.

  For example, the learning unit 13 adjusts the degree to which the dynamics held by the RNN is affected by the learning data to be small by decreasing the number of parameter calculation iterations as the learning weight is small.

  Alternatively, for example, the learning unit 13 adjusts the degree of the dynamics held by the RNN to be affected by the learning data to be small by correcting the prediction error to a smaller value as the learning weight is smaller.

  In any case, when the learning weight is large, the parameters of the RNN are updated so that the dynamics held by the RNN are greatly affected by the learning data. When the learning weight is small, the RNN parameters are updated so that the dynamics held by the RNN are not significantly affected by the learning data (only a little).

  Next, a method for determining the learning weight will be described.

  The learning unit 13 determines a winner node that holds the dynamics most suitable for the learning data from the nodes of the dynamics storage network, and holds each node according to the distance from the winner node to each node. The learning weight representing the degree to which the dynamics to be updated is determined is determined.

  That is, when a feature amount of one sample is supplied from the feature amount extraction unit 12, the learning unit 13 calculates the feature amount of the one sample and the past T-1 sample supplied immediately before from the feature amount extraction unit 12. Based on the feature amount, time-series data of the feature amount (sample value) of the T sample is generated as observed time-series data.

  Here, in this case, the time-series data obtained by sequentially extracting the T-samples while shifting the time-series data of the feature values output from the feature-value extraction unit 12 by one sample becomes the observation time-series data.

In the following, the (observation) time-series data X _t at time t is, for example, the sample value X _{t-T + 1} of the T sample from the sample value at time t-T + 1 to the sample value at time t. , X _{t−T + 2} ,..., X _t−1 , X _t .

  When learning unit 13 generates observation time-series data at time t, the learning time-series data at time t is used as learning data, and a score representing the degree to which each node (dynamics possessed by) of the dynamics storage network matches the learning data. Ask for.

That is, the time among the sample values X _{t−T + 1} , X _{t−T + 2} ,..., X _t−1 , X _t of time samples at time t as learning data When the sample value X _t of _t is input to the RNN as a dynamic system approximation model possessed by the node, the predicted value X ′ _{t + 1} of the sample value X _{t + 1} of the time t + 1 output by the RNN, time t + 1 sample value (true value) X _{t + 1} the prediction error [delta] with respect to (t) is, for example, the formula [delta] (t) = | defined by ² | X _{'t + 1} -X _{t + 1} If this is the case, the learning unit 13 predicts _T sample values X _{t-T + 1} , X _{t-T + 2} ,..., X _t−1 , X _t as time-series data at time t. For example, an addition value (sum) E _t (= δ (t) of errors δ (t−T + 1), δ (t−T + 2),..., Δ (t−1), δ (t) −T + 1) + δ (t−T + 2) +... + Δ (t−1) + δ (t)) is determined as the learning data for the time-series data at time t (whole) Find as a score.

  In this case, the smaller the score, the closer the predicted value is to the true value. Therefore, hereinafter, as appropriate, a score that is small is also referred to as a good or high score, and a score that is large is also referred to as a poor or low score.

  For a dynamical approximate model with an internal state quantity such as RNN, the score becomes better by setting the internal state quantity to an appropriate value.

  Therefore, in calculating the score, the learning unit 13 adjusts the RNN context as the internal state quantity by the BPTT method so as to minimize the prediction error, and then calculates the score while updating the context. .

  Then, the learning unit 13 determines a node having an RNN having the best score among the nodes of the dynamics storage network as a winner node that holds the dynamics most suitable for the learning data.

  Further, the learning unit 13 obtains a distance d between each node of the dynamics storage network and the winner node.

As the distance between the node N _i and N _j, other Euclidean distance between the node N _i and N _j described in FIG. 14, for example, the difference in scores between the nodes N _i and N _j ( (Absolute value) can be adopted. In this case, a node with a better score is a node closer to the winner node.

As the distance between any node N _i and winning node, it is possible to employ the score itself node N _i (or, reciprocal, etc.).

  When the learning unit 13 obtains the distance d between each node of the dynamics storage network and the winner node, a curve representing a relationship in which the learning weight α decreases with an increase in the distance d (hereinafter referred to as a distance / weight curve). ) To determine the learning weight α of the node.

  That is, FIG. 16 shows an example of a distance / weight curve.

  In the distance / weight curve of FIG. 16, the horizontal axis (from left to right) represents the learning weight α, and the vertical axis (from top to bottom) represents the distance d from the winner node.

  According to the distance / weight curve of FIG. 16, a node having a shorter distance d to the winner node determines a larger learning weight α, and a node having a longer distance d determines a smaller learning weight α.

Here, in FIG. 16, along the vertical axis, the six nodes N ₁ ′ to N ₆ ′ constituting the dynamics storage network correspond to the distance d between each node _Ni ′ and the winner node (vertical). Axis position).

In FIG. 16, the six nodes N ₁ ′ to N ₆ ′ constituting the dynamics storage network are nodes that are in the order of the distance d from the winner node. Among the _six nodes N ₁ ′ to N ₆ ′ constituting the dynamics storage network, the node N ₁ ′ having the closest distance d to the winner node, that is, the node N ₁ ′ having a distance of 0 from the winner node is It is the winner node.

  The distance / weight curve in FIG. 16 is given by, for example, Expression (1).

・・・（１）

... (1)

  Here, in Expression (1), γ is an attenuation coefficient in a range of 0 <γ <1, and Δ is a variable for adjusting the learning weight α of each node around the winner node (hereinafter, appropriately adjusted). Variable).

  As Δ is gradually approached to 0 from a large value, the learning weight α becomes a smaller value as the distance from the winner node increases. Basically, the adjustment variable Δ is adjusted so as to increase at the start of learning and to decrease with time.

  Based on the learning weight α in Expression (1), the parameter of the winner node (the parameter of the dynamic system approximation model that the node has) is updated so as to be most affected by the learning data, and the learning data increases as the distance from the winner node increases. The parameters of other nodes (nodes other than the winner node) are updated so as to reduce the influence of.

  Next, the learning process by the learning unit 13 in FIG. 13 will be described with reference to the flowchart in FIG.

  In step S11, the learning unit 13 initializes all parameters of the dynamics storage network stored in the network storage unit 16. Specifically, an appropriate value is assigned as an initial value to the parameter of the dynamic system approximation model (FIG. 15) having the internal state quantity of each node of the dynamics storage network.

  Here, when the dynamic system approximation model possessed by the node of the dynamics storage network is, for example, an RNN, in step S11, the coupling weight given to the signal input to the unit of the RNN is represented by the dynamic system approximation model. A suitable initial value is set as the parameter.

  Thereafter, after waiting for one sample of the feature amount to be supplied from the feature amount extraction unit 12 to the learning unit 13, the process proceeds from step S11 to step S12, and the learning unit 13 generates learning data. .

  That is, when a feature amount of one sample is supplied from the feature amount extraction unit 12, the learning unit 13 and the feature amount of the T-1 sample supplied immediately before from the feature amount extraction unit 12 As a result, time-series data of T sample feature values (sample values) is generated as observed time-series data.

  Thereafter, the process proceeds from step S12 to step S13, and the learning unit 13 uses the observation time series data generated in the immediately preceding step S12 as learning data, and the dynamics storage network stored in the network storage unit 16 for the learning data. The score of each node is calculated while updating the internal state quantity of the dynamic system approximation model having the internal state quantity of the node.

  Here, when the dynamic system approximation model having the internal state quantity is, for example, RNN, the score of the variable values whose values are changed (updated) based on a predetermined reference value is calculated. The best value is determined as the initial value of the RNN context as the internal state quantity, and the score is calculated while updating the context from the initial value.

  The predetermined reference value used for determining the initial value of the context is, for example, a random value or the final update value of the context obtained when the previous RNN parameter was updated (hereinafter, the previous update is appropriately performed as appropriate). Value)).

  For example, the observation time series data generated by the learning unit 13 when the RNN parameter is updated this time and the observation time series data generated by the learning unit 13 when the previous RNN parameter is updated have no relationship. If it is known, a random value can be adopted as the predetermined reference value used for determining the initial value of the context.

  Also, for example, when the observation time series data generated by the learning unit 13 at the time of the current RNN parameter update and the observation time series data generated by the learning unit 13 at the previous update of the RNN parameter are continuous. If it is known that there is some relationship, such as series data, the previous update value can be used as the predetermined reference value used to determine the initial value of the context. When the previous update value is adopted as a predetermined reference value used for determining the initial value of the context, the previous update value can be determined as the initial value of the context as it is.

  When the scores of all the nodes of the dynamics storage network stored in the network storage unit 16 are obtained, the process proceeds from step S13 to step S14, and the learning unit 13 compares the scores of the nodes constituting the dynamics storage network. As a result, the node having the best score is determined as the winner node that is the most suitable node for the learning data, and the process proceeds to step S15.

  In step S15, the learning unit 13 determines the learning weight of each node of the dynamics storage network stored in the network storage unit 16 with the winner node as the center, as described with reference to FIG.

  Thereafter, the process proceeds from step S15 to step S16, where the learning unit 13 uses the learning data and the dynamical system approximation model having the internal state quantity that each node of the dynamics storage network stored in the network storage unit 16 has. The parameter is updated according to the learning weight.

  Here, when the dynamic system approximation model having the internal state quantity is, for example, RNN, the parameter update in step S16 is performed by using the number of iterations for calculating the parameter (binding weight) by the BPTT method as the learning weight. Depending on the limit. That is, the smaller the learning weight, the smaller the number of repetitions.

  Note that the method of updating only the parameters of the winner node is a learning method called WTA (winner-take-all), and the method of updating parameters for nodes in the vicinity of the winner node is SMA (soft-max adaptation). ) Is a learning method. The learning unit 13 updates the parameters of the dynamics storage network (dynamic system approximation model possessed by the node) using SMA.

  That is, as described with reference to FIG. 16, the learning weight is determined to be larger for a node that is closer to the winner node and closer to the winner node, and conversely, for a node that is farther from the winner node. Decided to a small value. As a result, the node parameters are updated so that the nodes near the winner node are more affected by the learning data, and the nodes far from the winner node are less affected by the learning data. As described above, neighborhood competitive learning for updating the parameter of the node is performed.

  Thereafter, the process waits for the feature quantity of one sample to be newly supplied from the feature quantity extraction unit 12 to the learning unit 13, and the process returns from step S16 to step S12. Hereinafter, the processes of steps S12 to S16 are performed. Is repeated.

  Next, the recognition process by the recognition unit 14 in FIG. 13 will be described with reference to the flowchart in FIG.

  In step S31, the recognition unit 14 generates recognition data used for recognition processing.

  That is, for example, the recognition unit 14 waits for the feature amount (sample value) of the T sample to be supplied from the feature amount extraction unit 12, and obtains the observation time series data that is the time series of the feature amount of the T sample. And recognition data.

  Then, the process proceeds from step S31 to step S32, and the recognition unit 14 calculates the score of each node of the dynamics storage network for the recognition data, as in the learning process of FIG. This is done while updating the internal state quantity of the dynamic system approximation model with quantity.

  When the scores of all the nodes of the dynamics storage network are obtained, the process proceeds from step S32 to step S33, and the recognition unit 14 compares the scores of the respective nodes constituting the dynamics storage network to obtain the best score. The node is determined to be the winner node that is the most suitable node for the recognition data, and the process proceeds to step S34.

  In step S35, the recognition unit 14 outputs information representing the winner node as a recognition result of the recognition data, and the process ends.

  Here, the recognition result output by the recognition unit 14 can be output to the outside of the data processing apparatus of FIG. The recognition result output from the recognition unit 14 can be supplied to the generation unit 15 as a control signal.

  Next, generation processing by the generation unit 15 in FIG. 13 will be described with reference to the flowchart in FIG. 19.

  According to the learning process of FIG. 17, each node of the dynamics storage network learns and stores (acquires) the dynamics using a dynamical approximate model having an internal state quantity. Thereafter, the internal state quantity of each node is determined. It is possible to generate time series data having dynamics modeled by the dynamic system approximation model (time series data of a time series pattern acquired as dynamics) from the dynamic system approximation model.

  When RNN is used as a dynamical system approximation model with an internal state quantity, time series data can be easily generated from the dynamics held in the node having the RNN by giving the predetermined internal state quantity to the RNN. Can do.

  Specifically, when the state vector at the time t at the input of the RNN is given, the state vector at the next time t + 1 is output. Therefore, by performing this operation for a predetermined time step (sample point), time series data of the number of samples corresponding to the predetermined time step can be generated from each node of the dynamics storage network.

  That is, in step S51 of FIG. 19, the generation unit 15 determines which of the nodes corresponding to the dynamics among the nodes of the dynamics storage network is to generate the time series data.

  Here, a node used for generating time-series data is also referred to as a generation node as appropriate hereinafter. In the generation process, for example, the generation unit 15 randomly selects one node from the nodes of the dynamics storage network, and determines that node as a generation node. Or the production | generation part 15 determines the node used as a production | generation node from the nodes of a dynamics storage network based on the control signal supplied according to the instruction | indication from a user etc., for example.

  When the generation node is determined, the process proceeds from step S51 to step S52, and the generation unit 15 converts the time series data to the dynamic system based on the parameters of the dynamic system approximation model having the internal state quantity held by the generation node. The process proceeds to step S53 by generating the internal model while updating the internal state quantity of the approximate model.

  In step S53, the generation unit 15 converts time series data generated from the dynamic system approximation model of the generation node (hereinafter also referred to as generation time series data as appropriate) and outputs it as necessary, and the process ends. To do.

  Here, since the observation time series data as learning data used by the learning unit 13 for the learning process is a feature amount of the sensor motor signal, the generation time series data generated by the generation unit 15 is also the feature amount of the sensor motor signal. is there. The feature quantity of the sensor motor signal as the generation time series data is converted into a sensor motor signal by the generation unit 15 in step S53, and the motor signal of the sensor motor signal is supplied to, for example, an autonomous agent.

  When the dynamic system approximation model is, for example, RNN, when the generation time series data is generated by the generation unit 15, the initial value of the context input to the RNN context unit (FIG. 15) as an internal state quantity, For example, a random value is used as an initial value of data input to the input unit (FIG. 15).

  Also, as data input to the RNN input unit (FIG. 15) at a certain time t + 1, the predicted value of the data at time t + 1 output from the RNN output layer at the immediately previous time t is used. .

  Next, the recognition generation process by the recognition unit 14 and the generation unit 15 in FIG. 13 will be described with reference to the flowchart in FIG.

  As described above, according to the recognition generation process, the recognition behavior of the autonomous agent can be realized.

  When the recognition unit 14 and the generation unit 15 perform the recognition generation using the dynamics storage network in which the dynamics is learned by the dynamic system approximation model having the internal state quantity, the recognition processing in FIG. 18 and the generation processing in FIG. 19 are sequentially performed. It is difficult to generate a recognition that takes into account the internal state quantity of the dynamical system approximate model only by combining them.

  Therefore, the recognizing unit 14 and the generating unit 15 store the internal state quantity (internal state) of the dynamic system approximate model updated in the recognition process of FIG. 18 in the internal state storage unit 17, and display the internal state quantity. By using it in the generation process 19, the recognition generation process for generating the predicted value of the observation time series data at the next time t + 1 is performed on the observation time series data at the time t obtained from the observation signal.

  That is, in the recognition generation process, in step S71, the recognition unit 14 observes a time series of feature values (sample values) of T samples from the feature amount extraction unit 12 as in step S31 of FIG. Time series data is taken as recognition data.

  Thereafter, the process proceeds from step S71 to step S72, and the recognition unit 14 has a node that calculates the score of each node of the dynamics storage network for the recognition data, as in the learning process of FIG. This is done while updating the internal state quantity of the dynamical approximate model with the state quantity.

  However, in the calculation of the score in step S72, the recognition unit 14 reads the internal state quantity that has been updated and stored last time from the internal state storage unit 17, and uses the value read from the internal state storage unit 17 as a dynamical system approximation. The initial value of the model's internal state quantity (eg, RNN context).

  When the scores of all the nodes in the dynamics storage network are obtained, the process proceeds from step S72 to step S73, and the recognition unit 14 compares the scores of the respective nodes constituting the dynamics storage network to obtain the best score. The node is determined to be the winner node that is the node that best matches the recognition data.

  Further, in step S73, the recognition unit 14 updates the internal state quantity update value (updated internal state quantity) when the winner node is determined and the initial value of the internal state quantity when the winner node is determined. Are stored (stored) in the internal state storage unit 17.

  Here, the updated value of the internal state quantity stored in the internal state storage unit 17 is used as the internal state quantity (for example, the context of the RNN) of the dynamic system approximation model in step S72 in which the recognition unit 14 calculates the next score. ) Is used as the initial value.

  The initial value of the internal state quantity stored in the internal state storage unit 17 is used in the generation unit 15 when generating time-series data.

  Thereafter, the recognition unit 14 outputs information representing the winner node, and the process proceeds from step S73 to step S74. Information output from the recognition unit 14 is supplied to the generation unit 15 as a control signal.

  In step S74, the generation unit 15 has, as a generation node, a winner node represented by information supplied as a control signal from the recognition unit 14 among the nodes of the dynamics storage network, and has an internal state quantity held by the generation node. Based on the parameters of the dynamic system approximation model, generation time series data is generated while updating the internal state quantity of the dynamic system approximation model, and the process proceeds to step S75.

  That is, the generation unit 15 reads the stored value of the internal state storage unit 17 as an initial value of the internal state quantity of the dynamic system approximation model of the generation node of the dynamics storage network stored in the network storage unit 16.

  That is, the generation unit 15 reads the initial value of the internal state quantity when the generation node is determined as the winner node in the recognition unit 14 among the stored values of the internal state storage unit 17, and generates the dynamic system approximation model of the generation node Set to the initial value of the internal state quantity.

  Furthermore, the generation unit 15 generates the same recognition data that the recognition unit 14 generates in step S71 from the time series of the feature amounts supplied from the feature amount extraction unit 12, and the recognition data is generated as the generation node. The generation time series data is generated while giving the dynamic system approximation model and updating the internal state quantity of the dynamic system approximation model.

  Specifically, when the dynamic system approximation model is, for example, RNN, the generation node of the stored values of the internal state storage unit 17 is the winner in the recognition unit 14 with respect to the RNN context unit (FIG. 15). The initial value of the context when the node is determined is input as the initial value of the context when generating the generation time-series data.

  Furthermore, recognition data is input to the input unit of the RNN (FIG. 15).

  Then, while updating the internal state quantity of the dynamic system approximate model, generation time series data as a predicted value of observation time series data at the next time of observation time series data as recognition data is generated.

  In step S75, the generation unit 15 converts and outputs the generated time series data generated from the dynamic system approximate model of the generation node as necessary in the same manner as in step S53 in FIG. Returning to step S71, the processes of steps S71 to S75 are repeated.

  Here, the generation time series data generated by the generation unit 15 is the feature amount of the sensor motor signal, as described in FIG. 19, and the generation amount of the sensor motor signal feature amount is determined by the generation unit 15 in step S75. , Convert to sensor motor signal. And the motor signal of the sensor motor signal is supplied to an autonomous agent, for example.

  The robot performs the cognitive behavior by performing the recognition generation process in steps S71 to S75 in FIG. 20 as described above, for example, every hour.

  In the case of adopting a dynamics storage network as a dynamics learning model to be stored in the dynamics learning model storage unit 104 in the data processing apparatus of FIG. 1, the prediction learning unit 103 performs the learning process of FIG. In FIG. 20, by performing the recognition generation process of FIG. 20 as a prediction process, a large number of dynamics can be acquired, and a predicted value of desired operation data (and situation data) as time-series data having each of the large numbers of dynamics can be obtained. it can.

  Next, the series of processes described above can be performed by hardware or software. When a series of processing is performed by software, a program constituting the software is installed in a general-purpose computer or the like.

  Thus, FIG. 21 shows a configuration example of an embodiment of a computer in which a program for executing the series of processes described above is installed.

  The program can be recorded in advance on a hard disk 305 or a ROM 303 as a recording medium built in the computer.

  Alternatively, the program is stored temporarily on a removable recording medium 311 such as a flexible disk, a CD-ROM (Compact Disc Read Only Memory), an MO (Magneto Optical) disk, a DVD (Digital Versatile Disc), a magnetic disk, or a semiconductor memory. It can be stored permanently (recorded). Such a removable recording medium 311 can be provided as so-called package software.

  The program is installed in the computer from the removable recording medium 311 as described above, or transferred from the download site to the computer wirelessly via a digital satellite broadcasting artificial satellite, or a LAN (Local Area Network), The program can be transferred to a computer via a network such as the Internet. The computer can receive the program transferred in this way by the communication unit 308 and install it in the built-in hard disk 305.

  The computer includes a CPU (Central Processing Unit) 302. An input / output interface 310 is connected to the CPU 302 via the bus 301, and the CPU 302 is operated by an input unit 307 including a keyboard, a mouse, a microphone, and the like by the user via the input / output interface 310. When a command is input as a result of the equalization, a program stored in a ROM (Read Only Memory) 303 is executed accordingly. Alternatively, the CPU 302 also transfers a program stored in the hard disk 305, a program transferred from a satellite or a network, received by the communication unit 308 and installed in the hard disk 305, or a removable recording medium 311 attached to the drive 309. The program read and installed in the hard disk 305 is loaded into a RAM (Random Access Memory) 304 and executed. Thereby, the CPU 302 performs processing according to the above-described flowchart or processing performed by the configuration of the above-described block diagram. Then, the CPU 302 outputs the processing result from the output unit 306 configured with an LCD (Liquid Crystal Display), a speaker, or the like, for example, via the input / output interface 310, or from the communication unit 308 as necessary. Transmission and further recording on the hard disk 305 are performed.

  In this specification, the processing steps for describing a program for causing a computer to perform various types of processing do not necessarily have to be processed in chronological order according to the order described in the flowchart, and are executed in parallel or individually. Processing to be performed (for example, parallel processing or object processing) is also included.

  Further, the program may be processed by one computer or may be distributedly processed by a plurality of computers. Furthermore, the program may be transferred to a remote computer and executed.

  The embodiment of the present invention is not limited to the above-described embodiment, and various modifications can be made without departing from the gist of the present invention.

It is a block diagram which shows the structural example of 1st Embodiment of the data processor to which this invention is applied. It is a figure which shows the example of situation data and desired operation data. It is a flowchart explaining a learning process. It is a flowchart explaining a prediction process. It is a figure which shows the web browser on the screen of PC. It is a figure which shows the display screen of TV. It is a block diagram which shows the structural example of 2nd Embodiment of the data processor to which this invention is applied. It is a figure which shows a game screen. It is a block diagram showing a configuration example of a character action auxiliary module 152 _n. It is a figure which shows the example of situation data and desired operation data. It is a flowchart explaining a learning process. It is a flowchart explaining a prediction process. It is a block diagram which shows the structural example of one Embodiment of the data processor to which this invention is applied. It is a figure which shows typically the example of a dynamics storage network. It is a figure which shows the structural example of a node typically. It is a figure explaining the method of the determination of the learning weight in a learning process. It is a flowchart explaining a learning process. is there. It is a flowchart explaining a recognition process. It is a flowchart explaining a production | generation process. It is a flowchart explaining a recognition production | generation process. It is a block diagram which shows the structural example of one Embodiment of the computer to which this invention is applied.

Explanation of symbols

11 signal input unit, 12 feature quantity extraction unit, 13 learning unit, 14 recognition unit, 15
Generation unit, 16 network storage unit, 17 internal state storage unit, 101 status data acquisition unit, 102 operation data acquisition unit, 103 prediction learning unit, 104 dynamics learning model storage unit, 105 prediction unit, 106 operation data output unit, 151 teaching Character selection unit, 152 ₁ to 152 _N character motion assist module, 161 situation data acquisition unit, 162 operation data acquisition unit, 163 prediction learning unit, 164 dynamics learning model storage unit, 165 prediction unit, 166 operation data output unit, 301 bus , 302 CPU, 303 ROM, 304 RAM, 305 hard disk, 306 output unit, 307 input unit, 308 communication unit, 309 drive, 310 input / output interface, 311 removable recording medium

Claims (8)

In a data processing device that processes time-series data,
Situation data acquisition means for acquiring situation data which is time-series data representing the situation;
Operation data acquisition means for acquiring desired operation data which is time-series data corresponding to an operation desired by the user;
Learning means for learning the dynamics of the situation data and desired operation data;
Based on the dynamics, with the situation data as an input, prediction means for obtaining a predicted value of the desired operation data;
A data processing apparatus comprising: output means for outputting a predicted value of the desired operation data. The prediction means also obtains a predicted value of the situation data together with a predicted value of the desired operation data,
The data processing apparatus according to claim 1, wherein the output means outputs the predicted value of the desired operation data when a prediction error of the predicted value of the situation data with respect to a true value of the situation data is equal to or less than a predetermined threshold. . The data processing apparatus according to claim 1, wherein the learning unit learns the dynamics of the situation data and the desired operation data using a dynamics learning model that is a model capable of acquiring dynamics. The data processing according to claim 3, wherein the dynamics learning model is RNN (Recurrent Neural Network), FNN (Feed Forward Neural Network), SVR (Support Vector Regression), or RNN-PB (Recurrent Neural Net with Parametric Bias). apparatus. The data processing apparatus according to claim 3, wherein the dynamics learning model is a dynamics storage network configured by a plurality of nodes and holding dynamics in each of the plurality of nodes. The learning means updates the dynamics of each node of the dynamics storage network based on the situation data and desired operation data in a self-organizing manner,
The prediction means includes
Determining a winner node that is a node that holds the dynamics most suitable for the situation data;
The winner node is determined as a generation node which is a node used for generating time-series data,
The data processing device according to claim 5, wherein time-series data having dynamics held by the generation node is generated as a predicted value of the desired operation data. In a data processing method of a data processing apparatus for processing time series data,
While obtaining situation data that is time-series data representing the situation, obtain desired operation data that is time-series data corresponding to an operation desired by the user,
Learn the dynamics of the situation data and desired operation data,
Based on the dynamics, using the situation data as an input, obtain a predicted value of the desired operation data,
A data processing method including a step of outputting a predicted value of the desired operation data. In a program that causes a computer to function as a data processing device that processes time-series data,
Situation data acquisition means for acquiring situation data which is time-series data representing the situation;
Operation data acquisition means for acquiring desired operation data which is time-series data corresponding to an operation desired by the user;
Learning means for learning the dynamics of the situation data and desired operation data;
Based on the dynamics, with the situation data as an input, prediction means for obtaining a predicted value of the desired operation data;
A program that causes a computer to function as output means for outputting a predicted value of the desired operation data.

Images