JPH05265369A - Learning device - Google Patents

Abstract

PURPOSE:To improve learning effect more by deciding the mental state of a learner from the motion of the learner and displaying a question and an explanation corresponding to the mental state of the learner. CONSTITUTION:This device is equipped with a detecting means 1 which detects the motion of the learner, a deciding means 2 which decides the metal state of the learner from the detection result, a selecting means 3 which selects the proper question or explanation according to the decision result, a display means 5 which displays the selected question or explanation to the learner, and an input means 6 which inputs the answer of the learner to the selected question or explanation, and the selecting means 3 further selects whether the learner's answer inputted from the input means 6 is correct or not and selects a next question or explanation according to the mental state of the learner decided by the deciding means 2.

Description

Detailed Description of the Invention

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a learning device, and in particular,
The present invention relates to a computer learning device (CAI device).

[0002]

2. Description of the Related Art A CAI device stores a large number of problems with different degrees of difficulty and explanations of learning items in a database, and according to the degree of understanding of a learner, This is to present an appropriate question or explanation to the learner by selecting it. The learner answers the question presented. The CAI device presents a problem with higher difficulty to the learner if the learner's answer is correct. On the other hand, if the learner's answer is incorrect, the CAI device will present a less difficult question, or
Alternatively, the explanation of the learning items is displayed so that the learner can understand the learning items.

[0003]

In the conventional learning device as described above, the degree of understanding of the learning item is determined only by the answer of the learner to the presented question. Moreover, since the problem and the explanation were unilaterally presented regardless of the psychological state of the learner, the sufficient learning effect could not be obtained.

The learning device of the present invention has been made by paying attention to such a problem, and its purpose is to:
Judging the learner's psychological state based on the learner's movements,
An object of the present invention is to provide a learning device capable of improving the learning effect by presenting problems and explanations according to the psychological state of the learner.

[0005]

In order to achieve the above object, the learning device of the present invention comprises a detecting means for detecting a learner's motion, and a psychology of the learner based on the detection result from the detecting means. Judging means for judging the state, selecting means for selecting an appropriate problem or commentary based on the judgment from this judging means, and presenting means for presenting the problem or commentary selected by this selecting means to the learner, And an input means for inputting a learner's answer to the presented question or explanation,
The selecting means further selects the next question or explanation based on the correctness or incorrectness of the learner's answer inputted from the inputting means and the learner's psychological state judged by the judging means.

[0006]

That is, in the learning device of the present invention, the psychological state is determined from the learner's action, and an appropriate problem or explanation is selected based on the determined psychological state and presented to the learner. Alternatively, the next question or comment is further selected and presented to the learner based on the correctness of the learner's answer to the commentary and the psychological state of the learner.

[0007]

DESCRIPTION OF THE PREFERRED EMBODIMENTS An embodiment of the learning device of the present invention will be described in detail below with reference to the drawings. FIG. 1 shows C as an example.
The configuration of the AI device is shown, and the operation will be described below based on this configuration.

First, the learner's motion is continuously detected by the motion detector 1 including a television camera or a pressure sensor.
The detected time series signal is sent to the psychological state determination unit 2. The psychological state determination unit 2 determines, based on the learner's movement, the psychological state of the learner, for example, whether the learner is concentrated on learning or tired.

The question / explanation selection unit 3 selects an appropriate question or explanation from the database 4 based on the result of the determination and presents it to the learner on the display device 5. The learner inputs the answer using the answer input device 6. Problem / Explanation Selection Section 3
Selects the next question or explanation from the database 4 and presents it to the learner according to the correctness of the answer given by the learner and the psychological state of the learner judged by the psychological condition judging section 2. To do.

FIG. 2 is a block diagram of the motion detector 1 of FIG. This operation will be described below. The motion of the learner is detected by continuously photographing the face of the learner with the TV camera 10. The image taken by the TV camera 10 is read out at predetermined time intervals, converted into digital data by the A / D converter 11, and then stored in the memory A 12 or the memory B.
13 is stored. Memory A12 and memory B
The image data for one frame is alternately stored in 13 at read time intervals. That is, if the image is first stored in the memory A12, the next read image is stored in the memory B13, and the next image is stored again in the memory A1.
Stored in 2. Hereinafter, the memory A12 will be stored in the memory B13.
It is assumed that the image is stored after one read cycle.

As shown in FIG. 3A, the calculator 14 sets a window area of a predetermined size in the face image of the learner stored in the memory A12. Image memory A1
2 and the image stored in the image memory B13 are assigned addresses for each pixel in the horizontal and vertical directions with the upper left end of the image as the origin. Here, if the window size is H pixels in the horizontal direction and V pixels in the vertical direction, and the horizontal address of the upper left corner of the window is X and the vertical address is Y, the image in the window is expressed by the following formula. It is represented by the matrix W as in (1).

[0012]

[Equation 1] Here, W is scanned in the horizontal direction, and each scan line is connected from top to bottom to be vectorized to be f. That is,

[0013]

[Equation 2] Is. Each element of W and f is the brightness of the image.

Next, a window having the same size as the window set for the image in the memory A12 is set in the image stored in the memory B13. Then, the position of the origin of the window set in the memory B13 is changed horizontally and vertically within a predetermined range from the origin position (X, Y) of the window of the memory A12 as shown in FIG. The difference e _ij defined by the equation (3) is calculated to minimize i, j
Ask for.

[0015]

[Equation 3]

Here, h is the maximum number of pixels in the horizontal change of the origin position of the window set in the memory B13, and v
Is the maximum number of pixels in the vertical direction change of the origin position of the window set in the memory B13. Each of h and v is determined to be equal to or larger than the maximum value of moving horizontally and vertically in the image area of the TV camera 10 during the one read cycle.

G _ij is the window origin position (X + i,
The window image set in the memory B13 when Y + j) is represented by a vector as in the equation (2). Further, g _ij that minimizes the equation (3) is the memory A12.
It is the window image of the memory B13 that is the most similar to the image of the window part of. That is, i and j are 1
It represents the horizontal and vertical displacement of the face during a read cycle.

The displacement of the face image for each reading cycle in the horizontal and vertical directions calculated by the calculator 14 is stored in the memory 15 over a predetermined time interval. The horizontal and vertical displacements stored in the memory 15 (FIG. 4A) are
After the Fourier transform by the calculator 16, the square is calculated (FIG. 4B), and the squared values of the Fourier transform at a plurality of predetermined frequencies are extracted (see FIG. c)), and sent to the psychological state determination unit 2.

The psychological state determining section 2 is composed of a hierarchical neural network as shown in FIG. In this embodiment, a three-layer neural network is used, which includes an intermediate layer including the neuron units 21a to 21c, an output layer including the neuron units 22a to 22c, and a branch terminal 2.
0a to 20c constitute the input layer. Each neuron unit 21a-21c and 22a-22c performs the calculation of the following formula (4).

[0020]

[Equation 4]

Where X _i is the i-th input, W _i is the i-th input coupling strength, θ is the threshold of the neuron unit, N is the number of input signals, y is the output of the neuron unit, and f is It is a sigmoid function shown in the following equation (5).

[0022]

[Equation 5] FIG. 6 is a block diagram of each neuron unit.

The data of the value at the predetermined frequency of the square of the Fourier transform as shown in FIG. 4 (c) is sent from the arithmetic unit 16 to the psychological state judging section one after another. First, the first data is sent, branched at the branch terminals 20a to 20c, and input to the neuron units 21a to 21c.

In each of the neuro units 21a to 21c, the first data is input to the arithmetic unit 31 through the input buffer 30. In the arithmetic unit 31, this first data
The product of the data and the first weight data stored in the weight memory 35 is calculated and stored in the register 33.

Next, the second data is sent from the arithmetic unit 16 and branched at branching terminals 20a to 20c.
Input to the long units 21a to 21c. In each of the neuro units 21a to 21c, the second data is input to the arithmetic unit 31 through the input buffer 30. Calculator 31
Then, the product of this second data and the second weight data stored in the weight memory 35 is calculated, and the value of this product is added to the value stored in the register 33.

The same process is repeated for all the data sent from the arithmetic unit 16. After the calculation of all the data sent from the arithmetic unit 16 is completed, the arithmetic unit 31
Subtracts the value stored in the threshold memory 34 from the above calculation result stored in the register 33, and stores the resulting value in the function table 36 as x in the equation (5).
And f (x) are associated with each other, a value of f (x) corresponding to the value of the subtraction result is obtained as x and is output via the output buffer 32. In the weight memory 35, W in equation (4)
_i (i = 1 to N), and θ in the equation (4) is stored in advance in the threshold memory 34. These values are obtained by the learning device 24 by the generalized delta rule described later.

After the processing in the intermediate layer is completed, the outputs of the neuron units 21a to 21c in the intermediate layer are sequentially read from the neuron unit 21a, and output from the neurons in the output layer.
Input to the long units 22a to 22c. First of all,
The output of the long unit 21a is read. In each of the neuron units 22a to 22c in the output layer, the output data of the neuron unit 21a is input to the arithmetic unit 31 through the input buffer 30. The arithmetic unit 31 calculates the product of this data and the first data of the same number of weight data as the neuron unit of the intermediate layer stored in the weight memory 35, and stores it in the register 33. Remembered.

Next, the output of the neuron unit 21b is read out, and the neuron units 22a to 22 in the output layer are read out.
Enter in c. This second read data is input to the arithmetic unit 31 through the input buffer 30.

The arithmetic unit 31 calculates the product of this data and the second weight data stored in the weight memory, and the value of this product is added to the value stored in the register 33. It A similar process is repeated for the output data of all neuron units in the middle layer. After the calculation of the output data of all the neuron units in the intermediate layer is completed, the arithmetic unit 31 subtracts the value stored in the register 33, and the resulting value is stored in the function table 36. , The value of f (x) corresponding to the value of the result of the above subtraction is obtained as x from the correspondence table of x and f (x) in the equation (5),
Output through the output buffer 32.

Output unit neuron units 22a to 22
The pattern of the output value of c represents the psychological state of the learner. The data stored in the weight memory 35 and the threshold value memory 34 of each of the neuron units 22a to 22c of the output layer as well as the neuron units 21a to 21c of the intermediate layer.
The generalized delta rule devised by David Lamelhardt et al. L. McClellan, PDP Research Group, Shunichi Amari, translated, Sangyo Tosho (1989)).

First, the apparatus shown in FIG. 2 collects a large number of sample data relating to the learner's movement through the process described above and stores it in the sample data memory 23. The data stored in the sample data memory 23 is shown in FIG.
As shown in, the displacement pattern in the horizontal and vertical directions of the learner is Fourier transformed and squared, and is a set of values at a predetermined frequency. This set of values is used as the learning vector x _p (p = 1
~ M). m is the number of sample data. In the generalized delta rule,

[0032]

[Equation 6]

The weight and threshold value of each neuron unit are changed so as to decrease. Where y _p (p = 1,
…, M) is the learning vector x _p is the neural network.
This is the output pattern vector of the neural network obtained when input to the network. d _p is a learning vector x _p
On the other hand, it is a pattern vector that the neural network wants to output and is called a teacher vector. The teacher vector is input from the teacher signal input device 25 after the psychology expert determines the psychological state of the learner at the time of collecting the sample data. The update amount ΔW of each weight is obtained by the steepest descent method.

[0034]

[Equation 7] The change amount ΔW _ji of the i-th weight of the neuron unit j in the output layer is

[0035]

[Equation 8] And the change amount ΔW _ih of the h-th weight of the _neuron unit i in the middle layer is

[0036]

[Equation 9]

It can be obtained by x _hp is the h-th element of the learning vector x _p , and y _jp and y _ip are the output of the neuron unit j of the output layer when the learning vector x _p is input to the neural network, and This is the output value of the neuron unit i in the middle layer.

[0038] d _uk is the desired output pattern for learning vector x _p - a j th element of Nbekutoru d _p. r is the number of neuron units in the output layer. It should be noted that the threshold value θ can be obtained by the equations (8) and (9), assuming that −1 is always the weight of the terminal input to each neuron unit.

First, each weight is initialized by a random number having a small absolute value, and the equations (8) and (9) are repeatedly executed for each learning vector, and when E in the equation (6) becomes sufficiently small. Then, the processing ends with the weight memory 35 and the threshold memory 34 of the corresponding neuron unit for each weight value.
Remember. Through the above process, the psychological state determination unit 2 can determine the psychological state of the learner from the action of the learner.

The problem / commentary selection unit 3 is made into a database based on the patterns of the output values of the neuron units 22a to 22c in the output layer of the psychological state determination unit 2 which represent the psychological state of the learner. An appropriate one is selected from the stored problems and explanations and displayed on the display device 5. The learner inputs the answer to the question using the answer input device 6.

Further, the question / explanation selection unit 3 determines whether the learner's answer is correct or not, and is stored in the database based on the determination result and the output data of the psychological state determination unit. An appropriate problem is selected from the problems and explanations displayed and displayed to the learner by the display device 5. The above process is repeated during the learning time. It should be noted that the database 4, the game, the story, the music, etc. are stored in memory, and when the learner is determined to have come to study, he / she can select an appropriate one, and the learner It is also possible to be able to breathe out.

[0042]

According to the learning device of the present invention, the psychological state of the learner can be judged by the motion pattern of the learner, and problems and explanations corresponding to the psychological state of the learner can be presented. Is improved.

[Brief description of drawings]

FIG. 1 is a diagram showing a configuration of a computer learning device as an example of a learning device of the present invention.

FIG. 2 is a block diagram of a motion detector of FIG.

FIG. 3 is a diagram showing how a window area of a predetermined size is set in a face image of a learner.

FIG. 4 is a waveform diagram of a signal representing vertical and horizontal shifts of a learner.

5 is a diagram showing a configuration of a hierarchical neural network as a psychological state determination unit in FIG.

FIG. 6 is a block diagram of a neuron unit.

[Explanation of symbols]

DESCRIPTION OF SYMBOLS 1 ... Motion detection part, 2 ... Mental state determination part, 3 ... Question / commentary selection part, 4 ... Database, 5 ... Display device, 6 ... Answer input device.

Claims (1)

[Claims] 1. A detecting means for detecting a motion of a learner, a judging means for judging a psychological state of a learner based on a detection result from the detecting means, and an appropriate means based on the judgment from the judging means. It comprises a selection means for selecting a question or commentary, a presentation means for presenting the learner the question or commentary selected by this selection means, and an input means for inputting a learner's answer to the presented question or commentary. The selecting means further selects the next question or commentary based on the correctness of the learner's answer input from the inputting means and the psychological state of the learner determined by the determining means. And learning device.