CN113469213A

CN113469213A - Object identification method and device, terminal, processor and storage medium

Info

Publication number: CN113469213A
Application number: CN202110496365.8A
Authority: CN
Inventors: 熊楚渝
Original assignee: Chengdu Cyberkey Technologies Co ltd
Current assignee: Chengdu Cyberkey Technologies Co ltd
Priority date: 2017-04-28
Filing date: 2017-04-28
Publication date: 2021-10-01
Also published as: CN107194416A; CN107194416B

Abstract

The invention discloses an object identification method and device, a terminal, a processor and a storage medium. Wherein, the method comprises the following steps: receiving original characteristic data of an object to be identified; inputting the original feature data into a characterization space in a learning machine, and passing the original feature data through selected nodes in the characterization space to obtain intermediate data, wherein the nodes in the characterization space are connected through logical connection, and the nodes include at least one of the following: logic gates and transform functions; and identifying the intermediate data according to the selected prediction function to obtain an identification result of the object to be identified. The invention solves the technical problem that the learning machine in the related technology does not have the capability of supporting general mechanical learning.

Description

Object identification method and device, terminal, processor and storage medium

Technical Field

The invention relates to 2017102984812 divisional application, belongs to the field of machine learning, and particularly relates to an object identification method and device, a terminal, a processor and a storage medium.

Background

Machine learning is an important direction in the development of computing technology. It is well known that computers require manual programming of development software to enable them to do work. Machine learning has evolved into a new direction, namely to train a machine with data so that the machine can gain the ability to process this type of data through learning. By applying machine learning techniques, many software implementations may be developed that are not possible by programming alone, such as image recognition, language recognition, and so forth. At present, machine learning is a popular subject, and huge resources are invested in various aspects to carry out various researches. Thus, various aspects of machine learning have progressed rapidly. However, machine learning still faces numerous problems and dilemmas. One of the significant problems is that in fact, machine learning requires many manual interventions, and general mechanical learning cannot be achieved, where "general" means that a machine learning scheme is not only applicable to one type of situation, but also applicable to any type of situation; mechanical learning means that a machine follows its basic rules (also called meta-rules) and can improve itself according to its environment and input data without manual intervention, thereby further acquiring the ability to process relevant information and enabling normal operation without manual programming. However, the learning machine in the related art does not have a condition for realizing general mechanical learning. However, the learning machine in the related art does not have a condition for realizing the mechanical learning.

Disclosure of Invention

The embodiment of the invention provides an object identification method and device, a terminal, a processor and a storage medium, which are used for at least solving the technical problem that a learning machine in the related art does not have the capability of supporting general mechanical learning.

According to an aspect of an embodiment of the present invention, there is provided an object recognition method including: receiving original characteristic data of an object to be identified; inputting the original feature data into a characterization space in a learning machine, and passing the original feature data through selected nodes in the characterization space to obtain intermediate data, wherein the nodes in the characterization space are connected through logical connection, and the nodes include at least one of the following: logic gates and transform functions; and identifying the intermediate data according to the selected prediction function to obtain an identification result of the object to be identified.

Optionally, the transformation function comprises: and a group of nodes with connection relations.

Optionally, after inputting the raw feature data into a characterization space in a learning machine, the method further comprises: adjusting a learning method of a learning machine according to the intermediate data; and adjusting the characterization space and the prediction function according to the adjusted learning method.

Optionally, the adjusting the characterization space and the prediction function according to the adjusted learning method includes: adjusting at least one of the following elements in the characterization space according to the adjusted learning method: adding and deleting nodes, types of nodes, adding and deleting of the logical connection, types of the logical connection and credibility of the logical connection; and adjusting the prediction function according to the learning method.

Optionally, the states of the nodes include a connected state and a non-connected state, where the connected state refers to that the logical connection connected to the nodes other than the node exists on both the input side and the output side of the node, and the states of the nodes are determined according to the states of all the upstream nodes connected to the node and the states of the logical connection.

Optionally, the logical connection comprises at least one of the following types: the method comprises the steps of feedforward connection, connection forbidding, forced feedforward and feedforward forbidding, wherein the feedforward connection refers to the condition that data of a previous node is allowed to be received and input into a next node.

Optionally, identifying the intermediate data according to the selected prediction function to obtain an identification result of the object to be identified, including: determining an output node corresponding to the prediction function in the characterization space, and determining the state of the output node and the logical connection of the output node; and determining an output value of the prediction function according to the state of the output node and the type of the logic connection, and outputting the output value as the identification result.

According to another aspect of the embodiments of the present invention, there is also provided an object recognition apparatus, including: the receiving module is used for receiving the original characteristic data of the object to be identified; the processing module is used for inputting the original feature data into a characterization space in a learning machine, and passing the original feature data through selected nodes in the characterization space to obtain intermediate data, wherein the nodes in the characterization space are connected through logical connection, and the nodes include at least one of the following: logic gates and transform functions; and the identification module is used for identifying the intermediate data according to the selected prediction function to obtain an identification result of the object to be identified.

According to still another aspect of the embodiments of the present invention, there is also provided a storage medium including a stored program, wherein when the program runs, a device in which the storage medium is located is controlled to execute the object recognition method described above.

According to still another aspect of the embodiments of the present invention, there is also provided a processor for executing a program, where the program executes the object recognition method described above.

According to still another aspect of the embodiments of the present invention, there is further provided a processor, configured to run a learning machine, where the learning machine has a feature space, the feature space is composed of nodes, and each node is connected to another node through a logical connection, and the node includes at least one of: logic gates and transform functions.

Optionally, the learning machine further comprises the following modules: the learning module is connected with the representation space and used for providing a learning method and adjusting the learning method of the learning machine according to the intermediate data; adjusting the prediction functions corresponding to the representation space and the output space according to the adjusted learning method; the output space comprises one or more output nodes, and the output nodes are expressions and are used for obtaining output values according to a prediction function and feeding the output values back to the learning module so as to adjust the learning method.

According to another aspect of the embodiments of the present invention, there is also provided a terminal, including: the device comprises a first device, a second device and a third device, wherein the first device is used for acquiring original characteristic data of an object to be identified; the second device is used for inputting the original feature data into a characterization space in a learning machine and passing the original feature data through selected nodes in the characterization space to obtain intermediate data, wherein the nodes in the characterization space are connected through logical connection, and the nodes include at least one of the following: logic gates and transform functions; a processor that executes a program, wherein the program executes to execute the following processing steps for data output from the first and second devices: and identifying the intermediate data according to the selected prediction function to obtain an identification result of the object to be identified.

In the embodiment of the invention, the original characteristic data of the object to be recognized is processed by utilizing the representation space in the learning machine to obtain the intermediate data, and the representation space is formed by nodes formed by logic gates and/or transformation functions, so that the representation space can be adjusted, for example, the representation space is controlled and adjusted according to a learning method, powerful support is provided for mechanical learning, and the technical problem that the learning machine in the related technology does not have the capability of supporting general mechanical learning is solved.

Drawings

The accompanying drawings, which are included to provide a further understanding of the invention and are incorporated in and constitute a part of this application, illustrate embodiment(s) of the invention and together with the description serve to explain the invention without limiting the invention. In the drawings:

fig. 1 is a block diagram of a terminal according to an embodiment of the present invention;

FIG. 2 is a flow chart illustrating an object recognition method according to an embodiment of the present invention;

FIG. 3 is a schematic diagram of an alternative OSIPL learning machine according to an embodiment of the present invention;

FIG. 4 is a schematic flow diagram of the information flow of a learning machine according to an embodiment of the invention;

FIG. 5 is a schematic diagram of a basic structure of a token space according to an embodiment of the invention;

FIG. 6 is a schematic diagram of a node in a token space according to an embodiment of the present invention;

FIG. 7 is a schematic diagram of a connection in a token space according to an embodiment of the invention;

FIG. 8 is a logic diagram for characterizing a function of variation in space in accordance with an embodiment of the present invention;

FIG. 9 is a schematic diagram of an expression in a token space according to an embodiment of the invention;

FIG. 10 is a diagram illustrating an implementation principle of a prediction function according to an embodiment of the present invention;

fig. 11 is a flowchart of identity authentication feature extraction according to the related art;

FIG. 12 is a flow diagram of identity authentication feature extraction using a learning machine according to an embodiment of the present invention;

fig. 13a is a schematic diagram of a learning machine training in the internet of things according to an embodiment of the present invention;

fig. 13b is a schematic view of an application of the internet of things of a learning machine according to an embodiment of the invention;

fig. 14 is a block diagram of an object recognition apparatus according to an embodiment of the present invention.

Detailed Description

In order to make the technical solutions of the present invention better understood, the technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the drawings in the embodiments of the present invention, and it is obvious that the described embodiments are only a part of the embodiments of the present invention, and not all of the embodiments. All other embodiments, which can be derived by a person skilled in the art from the embodiments given herein without making any creative effort, shall fall within the protection scope of the present invention.

It should be noted that the terms "first," "second," and the like in the description and claims of the present invention and in the drawings described above are used for distinguishing between similar elements and not necessarily for describing a particular sequential or chronological order. It is to be understood that the data so used is interchangeable under appropriate circumstances such that the embodiments of the invention described herein are capable of operation in sequences other than those illustrated or described herein. Furthermore, the terms "comprises," "comprising," and "having," and any variations thereof, are intended to cover a non-exclusive inclusion, such that a process, method, system, article, or apparatus that comprises a list of steps or elements is not necessarily limited to those steps or elements expressly listed, but may include other steps or elements not expressly listed or inherent to such process, method, article, or apparatus.

According to an embodiment of the present invention, there is provided a terminal, as shown in fig. 1, including:

a first device 10 for acquiring original feature data of an object to be identified; optionally, the first device may be represented as an image capturing device, a sound capturing device, or the like, and may be flexibly selected according to actual situations. For example, when voice needs to be recognized, a voice collecting device provided in the terminal may be activated.

A second device 12, configured to input the raw feature data into a representation space in a learning machine, and pass the raw feature data through a selected node in the representation space to obtain intermediate data, where each node in the representation space is connected by a logical connection, and the node includes at least one of: logic gates and transform functions. Alternatively, the second device may be arranged to actively read the raw feature data in the first device, or may be arranged to read the raw feature data when a trigger condition is detected.

A processor 14, wherein the processor 14 executes a program, and wherein the program executes the following processing steps for data output from the first device and the second device: and identifying the intermediate data according to the selected prediction function to obtain an identification result of the object to be identified.

Based on the foregoing terminals, the embodiments of the present application provide an object recognition method, it should be noted that the steps shown in the flowchart of the drawings can be executed in a computer system such as a set of computer executable instructions, and although a logical order is shown in the flowchart, in some cases, the steps shown or described can be executed in an order different from that here. The method can be operated in an operating environment provided by the terminal shown in fig. 1, and fig. 2 is a schematic flow chart of an object identification method according to an embodiment of the present invention. As shown in fig. 2, the method comprises the following processing steps:

step S202, receiving original characteristic data of an object to be identified; alternatively, the raw feature data may be biometric data of a human body (e.g., voice information of a person, etc.), and for example, in an identification process, identification may be performed by using the collected biometric data.

Step S204, inputting the original feature data into a representation space in a learning machine, and passing the original feature data through a selected node in the representation space to obtain intermediate data, wherein each node in the representation space is connected through a logical connection, and the node includes at least one of the following: logic gates and transform functions. The above transformation function includes: and a group of nodes with connection relations.

It should be noted that, in an alternative embodiment, the learning machine may be an OSIPL learning machine, where the OSIPL indicates: an object diagram (Objective Patterns), Sampling (Sampling), Internal Representation space (Internal Representation), prediction functions (prediction functions), and Learning Methods (Learning Methods), where the object diagram is a Learning object. In order to enable the learning machine to be suitable for any learning task without manual setting, the internal representation space is an expression composed of logic gates and transformation functions, the expressions form nodes (namely transformation functions and logic gates) in the representation space, and a new representation space is formed based on the nodes, so that the representation space has any expansion and contraction capacity, can be expanded at will in learning and use, and does not need to be manually set with a structure in advance (all existing machine learning needs to be manually set with the structure), and therefore a continuously perfect expression can be formed for any learning object. On the internal characterization space, appropriate learning methods and a priori knowledge are matched, and the requirements of various mechanical learning can be met. The representation space is very different from the existing machine learning method, the representation space is no longer a black box, and the representation space can be completely transparent to the outside, so that the application field of the learning machine can be expanded.

And step S206, identifying the intermediate data according to the selected prediction function to obtain the identification result of the object to be identified.

In an optional embodiment, the learning machine according to this embodiment may be configured to not only recognize an object to be recognized, but also feed back an output result to the learning method portion (L) to adjust the learning method, specifically, after the original feature data is input into a representation space in the learning machine, adjust the learning method of the learning machine according to the intermediate data; and adjusting the characterization space and the prediction function according to the adjusted learning method.

Therefore, the learning method can be improved according to the output result, and the internal representation space can be adjusted.

In an optional embodiment, at least one of the following elements in the token space is adjusted according to the adjusted learning method: adding and deleting nodes, types of nodes, states of nodes, adding and deleting of the logical connections, types of the logical connections and credibility of the logical connections; and adjusting the prediction function according to the learning method. Wherein, the types of the nodes comprise: logic gates and transform functions. The state of the node includes a connected state and a non-connected state, wherein the connected state means that the logical connection connected to the node other than the node exists on both the input side and the output side of the node;

according to an alternative embodiment of the present application, the state of the node is determined according to the states of all upstream nodes connected to the node and the state of the logical connection.

Optionally, the logical connection includes at least one of the following types: feedforward coupling, inhibit coupling, force feedforward and inhibit feedforward, wherein:

the feedforward connection is that when the connected state of the current node is connected, the connected state of the next node is determined to be connected, and when the connected state of the current node is not connected, the connected state of the next node is determined to be not connected;

the connection prohibition means that when the state of the current node is connected, the connected state of the next node is determined to be connection prohibition, and when the state of the current node is non-connected, the connected state of the next node is determined to be non-connection prohibition;

the forced feed-forward means that when the state of the current node is connected, the connected state of the next node is determined to be disconnected, and when the state of the current node is disconnected, the connected state of the next node is determined to be connected;

the forced prohibition means that when the state of the current node is connected, the connected state of the next node is determined as non-prohibited connection, and when the state of the current node is non-connected, the connected state of the next node is determined as prohibited connection.

Optionally, the logical connection has the following properties: a confidence level, the confidence level taking on a real number between greater than 0 and less than 1. This confidence level is used to help the "learning method" (which may be thought of as a module or process node) determine how to learn and to help determine whether to connect the current logical connection.

Optionally, determining an output node corresponding to the prediction function in the characterization space, and determining a logical connection between a state of the output node and the output node; determining an output value of the prediction function according to a state of the output node and a kind of the logical connection, and outputting the output value as the recognition result.

The OSIPL learning machine will have the input space, output space, etc. that all IPUs have. However, the core components included in the OSIPL learning machine are not included in the general IP U. The IPU diagram can be seen in the literature l to learn the goal of the machine, which is to know from the input how to improve itself and further improve its ability to process information (note that the machine uses its word here, which is in line with the current academic world's knowledge of the machine). That is, there are several patterns in the input (patterns, which the learning machine can perceive to learn how to improve itself and further improve its ability to process them.

The learning machine according to the embodiment of the present application is described in detail below with reference to the drawings, and the following description is only an exemplary description and is not intended to limit the scope of the present invention.

The learning machine referred to below is a learning machine composed of O, S, I, P, L components, hereinafter referred to as an OSIPL learning machine or simply as a learning machine, which are independent from each other and cooperate with each other, so that the data processing efficiency of the learning machine can be improved considerably. This is different from various existing machine learning systems, such as various deep learning systems.

The OSIPL learning machine is an IPU (Internet protocol Unit) firstly, namely an information processor, and has information processing capability. However, not only the information processing capability but also the learning capability is provided, that is, learning from the environment and the input data is possible, and the information processing capability of the user is improved. Further, the OSIPL learning machine is also a general learning machine (the definition of the general learning machine, see document [ l ] >, that is, for any target schema (Objective scherns), the ability to process information of the target schema can be obtained through learning without programming.

It is noted here that the internal token space I is the core of the OSIPL learning machine, and that both processing information and learning are spread around I. The characterization space of the OSIPL is completely different from all current machine learning methods. Using this specially designed characterization space will give the OSIPL learning machine a lot of capabilities.

Fig. 3 is a schematic structural diagram of an alternative OSIPL learning machine according to an embodiment of the present invention. As shown in FIG. 3, the basic components of the learning machine are 1) a target schema (i.e., O, learning object); 2> sampling (i.e. S is the sampling process and method, the learning object, which must be sampled to act on the input space); 3> input space; 4) an internal characterization space (i.e., D, which is a core component of the learning machine, in which mechanical learning is embodied; 5> prediction function (i.e. P, which is a function that maps from the internal token space to the output space); 6) an output space; 7> learning method (i.e., L, which is a collection of various behavioral ways of controlling I). Note that S is actually a subset of O.

In an alternative embodiment, the basic principle of the OSIPL learning machine is as follows: when a base schema (an N-dimensional binary vector) enters the input internal token space, the token space performs information processing and learning actions, which include generating or modifying schema expressions (i.e., nodes in the internal token space) in the token space, selecting a prediction function to be used, and generating output values using the prediction function, where the output values are also a base schema and are M-dimensional binary vectors, and the behavior of the token space is controlled by a learning method in L, which includes generating and modifying expressions in the token space, and determining the prediction function, among other actions. After the output space has obtained the output value, it is possible to obtain feedback of this output value from the environment and can return this feedback to the learning machine, which L can obtain and can use this feedback to further modify the token space.

Fig. 4 is a schematic flow diagram of information flow of a learning machine according to an embodiment of the present invention, and as shown in fig. 4, in the OSIPL learning machine, there are two information flows, one is information processing, i.e., processing input information until an output is obtained. The other is to know how to improve oneself from the input information, and the various states of the existing I and L, and further improve oneself (this is learning). The two flows are parallel. It follows that the OSIPL learning machine will not be limited to performing so-called unsupervised learning or supervised learning, but can perform a variety of learning. Therefore, an important feature is obtained that the OSIPL learning machine can perform online learning without interruption. Of course, it is also allowed to interrupt learning.

It is often the case that L (i.e. the learning method) is unchanged. However, at a high level, the L itself will be embedded in another OSIPL learning machine, and in that case, the L itself will also learn. Such an OSIPL is called a level 2 learning machine and the functionality will be greatly improved. This is an important feature of the OSIPL learning machine, i.e., the capabilities and functions that may be implicit are made more explicit.

More specifically, the information processing capabilities of the OSIPL learning machine are embodied in connections and nodes in the token space, and learning is embodied in making appropriate changes to these connections and nodes, but such changes are not effected by external programming or the like, but rather by input of external data.

FIG. 5 is a schematic diagram of a basic structure of a token space according to an embodiment of the invention. As shown in fig. 5, the structure of the characterization space is as follows: the design of the characterization space in the middle of the OSIPL learning machine is the most important component in the present invention. A review of the characterization space used by other machine learning methods may be made herein. See document [2 ]. There, the characterization space is a network of artificial neurons, each of which is a linear combination and is further assigned a non-linear evaluation function. Furthermore, this token space works by manually setting the basic structure, e.g. how many neurons there are per layer, etc. Thus, the characterization space is eventually equivalent to a set of parameter spaces (a very voluminous parameter space). Such a space of characterization would have these significant limitations of 1) it is difficult to have a sharp characterization, i.e., it is very unclear what a large number of parameters are actually being expressed. Not only the meaning of these parameters cannot be obtained naturally, but also huge resources are invested to search for, and the meaning of the parameters is not necessarily clarified. In fact, this has become one of the major obstacles to the current deep learning. 2) After the basic structure is set manually, it is basically no longer possible to make large adjustments to the basic structure, thus making it unlikely to adapt to completely different situations, i.e. difficult to make a general-purpose learning machine. 3) Such artificial neuron structure determines that the learning method thereof needs to follow the existing mathematical method, especially the back propagation method, and is unlikely to adopt more updated methods. 4) These artificial neuron structures, which lack the ability and space to accommodate multiple learning methods, are not able to capture the benefits of multiple methods. The learning machine provided by the embodiment of the invention identifies the neurons as logic gates. However, there is no suitable solution for designing logic gates and connections in the characterization space in the related art. Therefore, a unique design is adopted in the representation space, so that the nodes in the representation space are specially designed logic gates or transformation functions (especially for geometric objects) for accomplishing special functions, and specially designed connections.

As shown in fig. 5, the basic units in the token space are nodes, which are logic gates or special function transformation functions. The nodes are connected by specially designed connections (as shown in fig. 7). The node may be a logic gate, or a transfer function with a special function controlled by L (see fig. 8). When the nodes are connected, an expression is formed. Each node is an expression (see fig. 9) formed by all nodes connected backwards, the representation space is controlled by L, and under the stimulation of S, the node can be generated, the connection can be modified (including the parameters of the connection), and the node can also disappear or lose the connection. Adding and reducing nodes in the representation space, and the like, are important links of a learning mechanism.

FIG. 6 is a diagram illustrating nodes in a token space according to an embodiment of the present invention. As shown in fig. 6, a node is a basic unit of the OSIPL learning machine to process information. Referring to fig. 4, the details of the nodes can be seen. The nodes are divided into two classes, one is a logic gate and the other is a transformation function owned by L. The OSIPL learning machine may have only logic gates and no transform functions, or both. In fact, the transformation function is a special set of pre-arranged logic gates and connections, pre-mastered by L, available without learning, and can be considered part of the a priori knowledge of the OSIPL learning machine. See fig. 6 for more details of the logic gates and the transformation functions. The logic gate or the transformation function is connected with 4 types: f. p, ff, fp, out of the rules of logic gates with 4 f, p, ff, fp, OSIPL connected: the logic gate has two states, light and black (1 and 0), and the connection (regardless of its kind) is also two states, light and black (1 and 0) if the logic gate is light, f, p are light, ff, fp are black in the outgoing connection, and if the logic gate is black, f, p are black, ff, fp are light in the outgoing connection. The rule of the transformation function of the OSIPL is completely formulated and standardized by the learning method, and can be actually obtained after a group of logic gates are specially connected.

If a node is a logic gate, the logic flow to determine that node can be seen in FIG. 6. if a node is a transform function, the logic flow is pre-customized.

A node may have incoming connections and outgoing connections. The node cannot control the state of the incoming connection, but the node decides the state of the outgoing connection, detailed in fig. 6.

It is often the case that a node may have both an incoming connection and an outgoing connection. However, a node may also have only incoming connections and no outgoing connections. It is also possible that a node has only outgoing connections but no incoming connections, and thus the node is intended to be in a black state, and the outgoing connections are all black, so this situation is virtually useless. A node may also be completely unconnected, but if so, the node is in a vanished state.

The OSIPL learning machine can add nodes or delete nodes. Adding and deleting nodes is controlled by the method of L. This is one of the basic mechanisms for learning by a learning machine.

FIG. 7 is a schematic diagram of a connection in a token space according to an embodiment of the invention. As shown in FIG. 7, the connections are also the basic units of the OSIPL learning machine to process information, and there are four total connections, feed forward connection, connection disabled, feed forward forced, and disable forced, which are f, p, ff, fp, respectively, in FIG. 7. Their properties are shown in figure 7. These connections are the basic mechanism by which the OSIPL learning machine processes information. However, any connection has not only the function of a connection but also a record of the trustworthiness of the connection. This record is a real number between 0 and 1. The credibility of each connection is controlled and adjusted by the method of L, which is an important link of a learning mechanism. In addition to the degree of trustworthiness being adjustable, the kind of connection between two nodes may be adjustable, e.g. from f to P, or from bite to impression, etc. The change of the kind of connection is controlled and adjusted by the method of L. This is also an important link of the learning mechanism. The connection between two nodes can be increased, deleted or even deleted completely. The addition and deletion of connections is controlled and regulated by the method of L. This is also an important link of the learning mechanism.

FIG. 9 is a diagram illustrating expressions in a token space, according to an embodiment of the present invention. Each node, together with all the nodes and all the connections below it, constitutes an expression in the token space, see fig. 9. This expression, I-represents the core content in space. These expressions are the basis for information processing. All nodes may have outgoing connections, so that the expression may become part of a higher level expression. Node B, as in fig. 9, is itself an expression, but is part of a higher level expression.

It should be noted that, the expressions in the representation space are all clear in logic and can be fully understood by people, so that the expressions can be completely transparent to the outside. Since the representation space is composed of various expressions, the representation space is completely transparent to the outside. This property is not available with existing machine learning methods. In many application areas, a characterization space that is transparent to the outside will be of paramount importance.

Fig. 10 is a schematic diagram illustrating an implementation principle of a prediction function according to an embodiment of the present invention. See fig. 10. The prediction function is a link of the output information of the OSIPL learning machine. After several nodes have been accumulated in the token space, the learning method will select one or more of the nodes (e.g., output node 1 in fig. 8, etc.), and then select the cells connected to the output space. The output space is an M-dimensional binary vector, and therefore requires M predictor function components (which together make up the predictor function). Thus, the output value of the prediction function is determined by the state and connection of the output nodes. If the connection is f, the state of the output node is equal to the output value of the prediction function. This is the usual case. Therefore, choosing the prediction function is substantially equal to choosing the output node.

It should be noted that the output node is an expression, and therefore, the selection of the prediction function is equivalent to the selection of an expression. This is the purpose of learning. The important step of learning of the OSIPL learning machine is to form an expression and select a proper expression. Basic contents of the learning method in the OSIPL learning machine L in the OSIPL is a part that integrates the learning method. The L part is separated from the other parts, so that greater flexibility and learning functions can be obtained. This is quite different from current machine learning systems. In current machine learning systems, the token space, expressions, confidence (if any), and learning methods are not clearly separated but intertwined, making it difficult to clearly process, and thus making improvements difficult. By adopting the scheme provided by the embodiment, the learning method and the characterization space are basically and completely separated. The learning method will have the basic functions of adding or deleting nodes, determining the type of nodes, adding or deleting connections, changing the type of connections, adjusting the reliability of connections, selecting prediction functions, etc. How the learning method implements these functions (e.g., whether it is deterministic or probabilistic, or a mixture of both, etc.) is not particularly critical to the invention. Any learning method having the above functions may be adopted. Of course, not all methods work equally, and it is the goal of further research to select a more efficient learning method.

It is generally prescribed that the learning method is invariant. However, the high level approach may also vary. For example, if an OSIPL learning machine is embedded in L itself, then L itself will evolve with learning, and this learning machine becomes a level 2 learning machine. When the OSIPL learning machine is provided with the above representation space and the learning method has the basic functions as described above, the OSIPL learning machine becomes a general learning machine. The selection of an excellent learning method can promote the improvement of the learning ability of the OSIPL.

The identity authentication is taken as an example for explanation

Identity authentication requires the use of advanced information processing techniques. Particularly, the identity authentication combined with biological characteristics requires data sampled in real time, and the data are subjected to deep processing, reliable characteristics are extracted from the data, and then judgment is made. Referring to fig. 11, one extremely disturbing problem is that in the conventional manner, these tasks are done by a manually programmed program. However, the programming of such a program requires a great labor input, and even after a great labor input, the effect is still very undesirable. This difficulty has been present. Various machine learning methods, such as a go game, are well developed. It is therefore desirable to overcome this difficulty with a method of machine learning, i.e., not by directly manually programming to the computing system, but by first establishing machine learning capabilities at the computing system and then making appropriate data inputs to the system so that the computing system can obtain the capabilities through learning. This is a very desirable direction to explore.

However, the current machine learning method is difficult to be used in the field of identity authentication. For more than one reason. But the impossibility can be seen only by looking at the computational resources required by these methods. For example, these methods often require hundreds of GPUs, require large amounts of data, and train for hundreds of hours. This is not conceivable for the field of identity authentication. Because the field of identity authentication needs to face hundreds of millions of users, it is not possible to allocate such an amount of computing resources to each user. See fig. 12, which is divided into 3 stages:

the first phase is the general training phase. At this stage, data is extracted for a certain population and trained with the data. For example, if the voice print feature is performed, the data of the Sichuan population or the data of the Shandong population may be selected. Training at this stage will enable the OSIPL learning machine to process the data for this population. Since all learning results of the OSIPL learning machine are reflected in the middle of its token space, this token space can be very easily used for the next stage of training. The data volume at this stage must be relatively large, and the training time may also be relatively long. However, the result can be used universally by a certain group of people.

The second phase is the personality training phase. At this stage, data of a certain person is taken for training. However, the initial token space is the result of the first stage of training, so that only a small amount of data (e.g., 20 fingerprint inputs, etc.) can be used, and thus the training time can be relatively short. At the end of this period, the OSIPL learning machine may have had a good ability to adapt to a person's data, specifically, error rates below a small threshold. Thus, after this phase is completed, its characterization space can be used for the use phase. It is important to note that the resources required at this stage are minimal. Can be completely finished in a short time by using a single current general-purpose CPU. This requirement for computing resources is critical to whether it can be practically deployed. Only very low requirements on computing resources must be made for use in the field of identity authentication.

The third stage is the actual use stage. This stage characterizes the training effort of the second stage in space. Thus, the calculation task of identity authentication can be accurately executed from the beginning. By the actual use stage, the OSIPL learning machine has the function of online learning, so online learning is continued. However, the learning result, i.e. the evolution of the characterization space, must be performed steadily under control, so that the safety of use can be ensured. In most cases, the characterization space will remain little or no variation. It must be noted that in both the second and third phases, an OSIPL learning machine is set for each user. This requires the OSIPL learning machine to use only small computing resources and be easily deployed. According to the understanding, the existing machine learning method cannot achieve the point. Each user is matched with a unique OSIPL learning machine, and the security is very excellent. This is also a security that cannot be achieved by all existing computing methods. Because the token space, which is essentially a refinement of many details of the user's identity, is essentially impossible to replicate (unless subjected to the exact same learning process, but this is simply impossible),

security is thus much more provided than in the prior art (i.e. storing features in a database).

The OSIPL learning machine may also be applied to other application scenarios: the use of the OSIPL learner in the field of identity authentication is discussed previously. Although this is only an application in one field, the same J concept can be applied in many different fields. For example, an intelligent node in the internet of things, an OSIPL learning machine, may be well applicable.

Referring to fig. 13a-13b, this is a scenario of heavy usage of intelligent nodes in a local part of the internet of things. Because the internet of things requires the nodes to be very energy-saving, online learning can be achieved, and the cost is very low, the intelligent nodes of the internet of things can be applied only when the intelligent nodes meet the requirements. There is also a more important requirement that manual programming and extensive manual management must be avoided as much as possible. The OSIPL learning machine, as per fig. 13a-13b, will fully satisfy these conditions. Fig. 13b illustrates the deployment of the intelligent nodes. During the use process, the nodes receive a great deal of information, which may be information from other nodes, information from a server or other information sources, and after receiving the information, intelligent judgment and action are required to be made, including reporting to the server and the like. If the node's ability to process this information comes from manual programming, it can be difficult to manage because of the sheer number of nodes and the need to change often to cope with rapidly changing conditions. For example, if the network is the internet of things that detects vehicle traffic, some nodes, such as node 3 and node M in fig. 13a-13b, need to report an abnormal situation to the server. If the software of the nodes is completely developed by manual programming, the nodes are difficult to adapt to the rapidly changing conditions and make accurate judgment on abnormal conditions, so that huge manpower is required for maintaining the nodes, but the nodes are difficult to maintain. Therefore, it is advantageous that the nodes have the ability to learn from the input information, which greatly reduces the management effort. Fig. 13a shows a process of capability acquisition of an intelligent node. This is the same as before, with general training, and the individual using two parts of online learning. It is because the OSIPL learning machine is employed, and both of these parts, that deployment as shown in fig. 13b is possible.

Fig. 14 is a block diagram of an object recognition apparatus according to an embodiment of the present invention. As shown in fig. 14, the apparatus includes: a receiving module 140, configured to receive original feature data of an object to be identified; the processing module 142 is configured to input the raw feature data into a characterization space in a learning machine, and pass the raw feature data through a selected node in the characterization space to obtain intermediate data, where each node in the characterization space is connected by a logical connection, and the node includes at least one of: logic gates and transform functions; and the identification module 144 is configured to identify the intermediate data according to the selected prediction function to obtain an identification result of the object to be identified.

It should be noted that, the above modules may be implemented in software or hardware, and for the latter, the following may be implemented, but not limited to: the modules are positioned in the same processor; alternatively, the modules may be located in different processors.

The embodiment of the invention provides a storage medium, which comprises a stored program, wherein when the program runs, the device where the storage medium is located is controlled to execute the object identification method.

The embodiment of the invention provides a processor, which is used for running a program, wherein the program executes the object identification method when running.

The embodiment of the invention also provides a terminal, the learning machine is provided with a representation space, the representation space is composed of nodes, all the nodes are connected through logic connection, and the nodes comprise at least one of the following components: logic gates and transform functions.

The learning machine further comprises the following modules: the learning module is connected with the representation space and used for providing a learning method and adjusting the learning method of the learning machine according to the intermediate data; adjusting the prediction functions corresponding to the representation space and the output space according to the adjusted learning method; the output space comprises one or more output nodes, and the output nodes are expressions and are used for obtaining output values according to a prediction function and feeding the output values back to the learning module so as to adjust the learning method.

The above-mentioned serial numbers of the embodiments of the present invention are merely for description and do not represent the merits of the embodiments.

In the above embodiments of the present invention, the descriptions of the respective embodiments have respective emphasis, and for parts that are not described in detail in a certain embodiment, reference may be made to related descriptions of other embodiments.

In the embodiments provided in the present application, it should be understood that the disclosed technology can be implemented in other ways. The above-described embodiments of the apparatus are merely illustrative, and for example, the division of the units may be a logical division, and in actual implementation, there may be another division, for example, multiple units or components may be combined or integrated into another system, or some features may be omitted, or not executed. In addition, the shown or discussed mutual coupling or direct coupling or communication connection may be an indirect coupling or communication connection through some interfaces, units or modules, and may be in an electrical or other form.

The units described as separate parts may or may not be physically separate, and parts displayed as units may or may not be physical units, may be located in one place, or may be distributed on a plurality of units. Some or all of the units can be selected according to actual needs to achieve the purpose of the solution of the embodiment.

In addition, functional units in the embodiments of the present invention may be integrated into one processing unit, or each unit may exist alone physically, or two or more units are integrated into one unit. The integrated unit can be realized in a form of hardware, and can also be realized in a form of a software functional unit.

The integrated unit, if implemented in the form of a software functional unit and sold or used as a stand-alone product, may be stored in a computer readable storage medium. Based on such understanding, the technical solution of the present invention may be embodied in the form of a software product, which is stored in a storage medium and includes instructions for causing a computer device (which may be a personal computer, a server, or a network device) to execute all or part of the steps of the method according to the embodiments of the present invention. And the aforementioned storage medium includes: a U-disk, a Read-Only Memory (ROM), a Random Access Memory (RAM), a removable hard disk, a magnetic or optical disk, and other various media capable of storing program codes. The foregoing is only a preferred embodiment of the present invention, and it should be noted that, for those skilled in the art, various modifications and decorations can be made without departing from the principle of the present invention, and these modifications and decorations should also be regarded as the protection scope of the present invention.

Claims

1. An object recognition method, comprising:

receiving original characteristic data of an object to be identified;

inputting the original feature data into a characterization space in a learning machine, and passing the original feature data through selected nodes in the characterization space to obtain intermediate data, wherein the nodes in the characterization space are connected through logical connection, and the nodes include at least one of the following: logic gates and transform functions;

and identifying the intermediate data according to the selected prediction function to obtain an identification result of the object to be identified.

2. The method of claim 1, wherein the transformation function comprises: and a group of nodes with connection relations.

3. The method of claim 1, wherein after inputting the raw feature data into a characterization space in a learning machine, the method further comprises:

adjusting a learning method of a learning machine according to the intermediate data; and

and adjusting the characterization space and the prediction function according to the adjusted learning method.

4. The method of claim 1, wherein adjusting the characterization space and the prediction function according to the adjusted learning method comprises:

adjusting at least one of the following elements in the characterization space according to the adjusted learning method: adding and deleting nodes, types of nodes, states of nodes, adding and deleting of the logical connection, types of the logical connection and credibility of the logical connection; and adjusting the prediction function according to the learning method.

5. The method of claim 1,

the states of the nodes comprise a connected state and a non-connected state, wherein the connected state refers to the condition that the logic connection connected with the nodes except the nodes exists on the input side and the output side of the nodes; and/or

The state of the node is determined in dependence on the states of all upstream nodes connected to the node and the state of the logical connection.

6. The method according to any of claims 1 to 5, wherein the logical connection comprises at least one of the following types: feedforward connection, inhibition connection, forced feedforward and inhibition feedforward;

the connection prohibition means that when the state of the current node is connected, the connected state of the next node is determined to be connection prohibition, and when the state of the current node is disconnected, the connected state of the next node is determined to be non-connection prohibition;

the forced feedforward is to determine that the connected state of the next node is non-connected when the state of the current node is connected, and determine that the connected state of the next node is connected when the state of the current node is non-connected;

the forced prohibition means that when the state of the current node is connected, the connected state of the next node is determined to be non-prohibited connection, and when the state of the current node is non-connected, the connected state of the next node is determined to be prohibited connection.

7. The method according to claim 6, wherein identifying the intermediate data according to the selected prediction function to obtain the identification result of the object to be identified comprises:

determining an output node corresponding to the prediction function in the characterization space, and determining the state of the output node and the logical connection of the output node;

and determining an output value of the prediction function according to the state of the output node and the type of the logic connection, and outputting the output value as the identification result.

8. The method according to any of claims 1 to 5, wherein the logical connection has the following properties: a confidence level, the confidence level taking on a real number between greater than 0 and less than 1.

9. An object recognition apparatus, comprising:

the receiving module is used for receiving the original characteristic data of the object to be identified;

the processing module is used for inputting the original feature data into a characterization space in a learning machine, and passing the original feature data through selected nodes in the characterization space to obtain intermediate data, wherein the nodes in the characterization space are connected through logical connection, and the nodes include at least one of the following: logic gates and transform functions;

and the identification module is used for identifying the intermediate data according to the selected prediction function to obtain an identification result of the object to be identified.

10. A storage medium, characterized in that the storage medium comprises a stored program, wherein when the program runs, a device in which the storage medium is located is controlled to execute the object recognition method according to any one of claims 1 to 8.

11. A processor configured to execute a program, wherein the program executes to perform the object recognition method according to any one of claims 1 to 8.

12. A terminal, wherein the terminal provides a runtime environment for a learning machine, the learning machine having a representation space, the representation space comprising nodes, each of the nodes being connected by a logical connection, and wherein the nodes comprise at least one of: logic gates and transform functions.

13. The terminal of claim 12, wherein the learning machine further comprises the following modules:

the learning module is connected with the representation space and used for providing a learning method and adjusting the learning method of the learning machine according to the intermediate data; adjusting the prediction functions corresponding to the representation space and the output space according to the adjusted learning method;

the output space comprises one or more output nodes, and the output nodes are expressions and are used for obtaining output values according to a prediction function and feeding the output values back to the learning module so as to adjust the learning method.

14. A terminal, comprising:

the device comprises a first device, a second device and a third device, wherein the first device is used for acquiring original characteristic data of an object to be identified;

the second device is used for inputting the original feature data into a characterization space in a learning machine and passing the original feature data through selected nodes in the characterization space to obtain intermediate data, wherein the nodes in the characterization space are connected through logical connection, and the nodes include at least one of the following: logic gates and transform functions;

a processor that executes a program, wherein the program executes to execute the following processing steps for data output from the first and second devices: and identifying the intermediate data according to the selected prediction function to obtain an identification result of the object to be identified.