Symbol reasoning method based on fusion of symbolic logic and brain-like neural network

Abstract

The invention relates to the technical field of computers, and discloses a symbolic reasoning method based on fusion of symbolic logic and a brain-like neural network, comprising the following steps: collecting environmental information; mapping into a symbol pulse time sequence of the neuron population by using neuron population coding representation; the pulse neural network is used for processing, synaptic connection between the neuron groups is activated, the discharge time sequence of the neuron groups is output, and logic relation symbol reasoning between different symbols is established; adjusting the synaptic connection strength among the neuron groups by using an STDP learning mechanism, and outputting pulses generated by interaction of the neuron groups; in combination with symbol logic rules and symbol logic reasoning, distributed and robust representation is provided for symbols through neuron group coding, a symbol pulse time sequence is processed by using a pulse neural network, a symbol logic reasoning process is simulated by using an STDP learning mechanism, and symbol logic reasoning is performed autonomously.

Description

Technical Field

[0001] The present invention relates to the field of computer technology, and in particular to a symbolic reasoning method based on the integration of symbolic logic and brain-like neural networks. Background Art

[0002] In the field of artificial intelligence, symbolic reasoning and neural networks are two major intelligent computing methods, each with its own unique advantages and limitations. Symbolic reasoning relies on clear logical rules and symbolic representations, and can perform precise logical operations and reasoning, but it has limitations in processing unstructured data and complex pattern recognition. Neural networks, especially spiking neural networks (SNNs), have made significant progress in image recognition, speech processing and other fields with their powerful nonlinear fitting and pattern recognition capabilities. However, SNNs are relatively weak in symbolic reasoning and precise logical operations.

[0003] In the prior art, some studies have combined symbolic reasoning with neural networks, for example, embedding symbolic information into the input layer of the neural network so that the network can learn symbolic rules, or building a hybrid model that combines symbolic reasoning modules with neural network modules to achieve more complex reasoning tasks. However, these methods usually have some problems, such as low learning efficiency, limited reasoning ability, and difficulty in adapting to dynamically changing environments. Summary of the invention

[0004] In view of this, the present invention provides a symbolic reasoning method based on the integration of symbolic logic and brain-like neural network to solve the limitations of the prior art such as low learning efficiency.

[0005] In a first aspect, the present invention provides a symbolic reasoning method based on the integration of symbolic logic and brain-like neural network, the method comprising: Collect environmental information; The neuron population is used to encode environmental information and map it into a symbolic pulse time sequence of the neuron population. Different neuron populations are connected through synapses, and neurons produce synchronized pulse discharge patterns. Use spiking neural networks to process symbolic pulse time series, activate synaptic connections between neuron groups, output the discharge timing of each neuron group, establish logical relationships between different symbols, and perform symbolic reasoning; The STDP learning mechanism is used to adjust the synaptic connection strength between neuron groups based on the discharge timing of each neuron group, and output the pulses generated by the interaction of neuron groups; According to the pulses generated by the interaction of neuronal groups, combined with symbolic logic rules, symbolic logic reasoning is performed to obtain symbolic logic reasoning results.

[0006] The present invention uses neuron group encoding to represent the collected environmental information, maps it into the pulse time series of the neuron group, provides a distributed and robust representation for symbols, uses a pulse neural network to process the symbol pulse time series, provides a brain-like mechanism for symbol representation and operation, and uses the STDP learning mechanism to simulate the symbolic reasoning process, without the need to display complex coding rules, to achieve autonomous learning, combines symbolic logic rules, performs symbolic logic reasoning, and autonomously completes generalized reasoning, which greatly improves the intelligence level and also injects adaptability and robustness into symbolic logic reasoning.

[0007] In an optional implementation, after collecting the environmental information, the method further includes: Information is extracted from environmental information and represented in the form of triples, which are composed of entities and relation words.

[0008] The present invention expresses the information extracted from the environmental information in the form of triples, thereby facilitating sending the data in the form of triples for subsequent processing.

[0009] In an optional implementation, the environmental information is represented by encoding a neuron population and mapped into a symbolic pulse time sequence of the neuron population, including: Assign neuron populations to entities in the triplet respectively, wherein neuron populations of different entities are connected through synapses; The time delay encoding of the relational words in the triplet results in a specific time delay between the firing of the neuronal population; The specific time delays between neuronal groups and neuronal group discharges are mapped to symbolic spike time sequences of the neuronal groups, which are suitable for SNN processing.

[0010] The present invention implements distributed neural representation of symbols by allocating neuron groups to entities in triples, encoding relational words in triples, and using neuron groups to perform symbol encoding.

[0011] In an optional embodiment, the spiking neural network includes bottom-layer neurons, middle-layer neurons, and top-layer neurons. The spiking neural network is used to process the symbol pulse time sequence, activate the synaptic connection between the neuron groups, output the discharge timing of each neuron group, establish the logical relationship between different symbols, and perform symbolic reasoning, including: The underlying neurons are used to encode individual symbols in the symbol spike time series and activate synaptic connections between neuron populations; The middle layer neurons are used to establish the logical relationship between different symbols in the symbol pulse time series according to the specific time delay between the firing of the neuron group; The top layer of neurons is used to perform symbolic reasoning based on the logical relationship between different symbols and combined with contextual information.

[0012] The present invention processes the pulse time series through the bottom-layer neurons, middle-layer neurons and top-layer neurons in the pulse neural network, captures the logical relationship between symbols, performs symbolic reasoning, and injects adaptability and robustness into symbolic reasoning.

[0013] In an optional embodiment, adjusting the synaptic connection strength between neuron groups based on the discharge timing of each neuron group includes: If the order in which the neuron groups produce discharges is consistent with the discharge timing, the strength of the synaptic connection between the neuron groups is strengthened; If the order in which neuronal groups generate discharges is inconsistent with the discharge timing, the strength of the synaptic connection between neuronal groups will be weakened.

[0014] The present invention adjusts the synaptic connection strength between neuron groups so that the neuron groups can correctly reflect the input symbolic logic knowledge.

[0015] In an optional implementation, symbolic logic reasoning is performed based on the pulses generated by the interaction of the neuron population in combination with symbolic logic rules to obtain symbolic logic reasoning results, including: Decoding the pulses generated by the interaction of neuronal groups to obtain symbolic concepts; Symbolic logic reasoning is performed based on symbolic concepts using a pre-established semantic knowledge base that stores symbolic logic rules.

[0016] The present invention combines the pulse outputs generated by the interaction of neuron groups with existing symbolic logic rules to perform precise logical operations and reasoning, effectively overcoming the lack of interpretability of deep neural networks, improving interpretability and transparency, and helping to enhance the efficiency of human-computer collaboration.

[0017] In a second aspect, the present invention provides a symbolic reasoning system based on the integration of symbolic logic and brain-like neural networks, the system comprising: Perception module, used to collect environmental information; The symbol encoding module is used to encode environmental information using neuron groups and map it into a symbol pulse time sequence of the neuron groups. Different neuron groups are connected through synapses, and neurons generate synchronous pulse discharge patterns. The neuron group interaction module is used to process the symbol pulse time series using the spiking neural network, activate the synaptic connection between neuron groups, output the discharge timing of each neuron group, establish the logical relationship between different symbols, and perform symbolic reasoning; STDP learning module, used to adjust the synaptic connection strength between neuron groups based on the discharge timing of each neuron group using the STDP learning mechanism, and output the pulses generated by the interaction of the neuron groups; The symbolic logic reasoning module is used to perform symbolic logic reasoning based on the pulses generated by the interaction of the neuron group and in combination with the symbolic logic rules to obtain the symbolic logic reasoning results.

[0018] In a third aspect, the present invention provides a computer device comprising: a memory and a processor, the memory and the processor being communicatively connected to each other, the memory storing computer instructions, and the processor executing the symbolic reasoning method based on the fusion of symbolic logic and brain-like neural network of the above-mentioned first aspect or any corresponding embodiment thereof by executing the computer instructions.

[0019] In a fourth aspect, the present invention provides a computer-readable storage medium having computer instructions stored thereon, wherein the computer instructions are used to enable a computer to execute the symbolic reasoning method based on the fusion of symbolic logic and brain-like neural network according to the above-mentioned first aspect or any corresponding embodiment thereof.

[0020] In a fifth aspect, the present invention provides a computer program product, comprising computer instructions, which are used to enable a computer to execute the symbolic reasoning method based on the fusion of symbolic logic and brain-like neural network according to the above-mentioned first aspect or any corresponding embodiment. BRIEF DESCRIPTION OF THE DRAWINGS

[0021] In order to more clearly illustrate the specific implementation methods of the present invention or the technical solutions in the prior art, the drawings required for use in the specific implementation methods or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are some implementation methods of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0022] Figure 1 is a schematic diagram of the overall architecture of a symbolic reasoning system based on the integration of symbolic logic and brain-like neural network according to an embodiment of the present invention; Figure 2 It is a flowchart of a symbolic reasoning method based on the integration of symbolic logic and brain-like neural network according to an embodiment of the present invention; Figure 3 is a schematic diagram of the working process of a symbol encoding module according to an embodiment of the present invention; Figure 4 is a schematic diagram of a symbolic logic reasoning process according to an embodiment of the present invention; Figure 5is a structural block diagram of a symbolic reasoning system based on the integration of symbolic logic and brain-like neural network according to an embodiment of the present invention; Figure 6 It is a schematic diagram of the hardware structure of a computer device according to an embodiment of the present invention. DETAILED DESCRIPTION

[0023] In order to make the purpose, technical solution and advantages of the embodiments of the present invention clearer, the technical solution in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative work are within the scope of protection of the present invention.

[0024] In order to combine the accuracy of symbolic reasoning with the generalization ability of neural networks, the embodiment of the present invention provides a symbolic reasoning method based on the integration of symbolic logic and brain-like neural networks. By means of neuron population encoding, symbolic logic and brain-like pulse neural networks are combined to achieve efficient symbol representation and logical reasoning. The present invention aims to solve the gap between symbolic reasoning and neural networks in the prior art, realize the complementary advantages of the two, and promote the development of artificial intelligence technology.

[0025] According to an embodiment of the present invention, an embodiment of a symbolic reasoning method based on the fusion of symbolic logic and brain-like neural network is provided. It should be noted that the steps shown in the flowchart of the accompanying drawings can be executed in a computer system such as a set of computer executable instructions, and although the logical order is shown in the flowchart, in some cases, the steps shown or described can be executed in an order different from that shown here.

[0026] In this embodiment, a symbolic reasoning method based on the integration of symbolic logic and brain-like neural network is provided, which is applied to a symbolic reasoning system based on the integration of symbolic logic and brain-like neural network. The system consists of five main modules, such as Figure 1 As shown in the figure, it includes perception module (knowledge extraction), symbol encoding module, neuron group interaction module, STDP learning module and symbolic logic reasoning module. There is information flow between each module, forming a closed-loop decision-making control system.

[0027] Figure 2 is a flowchart of a symbolic reasoning method based on the fusion of symbolic logic and brain-like neural network according to an embodiment of the present invention, such as Figure 2 As shown, the process includes the following steps: Step S201, collecting environmental information.

[0028] In the embodiment of the present invention, the perception module is responsible for collecting environmental information and inputting data for the decision-making system, including multimodal information such as sensor data, images and videos. Common perception devices include sensors, image cameras, microphones, etc. The collected environmental information includes the robot's position, speed, distance to objects, and multimodal information such as vision and hearing.

[0029] The core of the perception module is to convert external input into symbolic logical assertions to be reasoned, that is, to convert it into a structured data form that can be accepted by the symbolic encoding module.

[0030] Step S202, using neuron group coding to represent environmental information, and mapping it into a symbolic pulse time sequence of the neuron group.

[0031] In an embodiment of the present invention, the symbol encoding module converts the environmental information from the perception module into a symbol pulse time sequence suitable for SNN processing. The symbol encoding module adopts a symbol group encoding algorithm to efficiently encode the symbolic logic information and efficiently maps the symbolic logic information into a pulse time sequence of a neuron group.

[0032] The neuron populations corresponding to different symbols are linked to each other through synaptic connections, which reflect the knowledge of the logical relationship between the symbols. For example, there is a synaptic connection between the neuron populations corresponding to the two symbols "dog" and "animal", indicating that "dog" is an instance of "animal".

[0033] For example, the concept of "dog" can be encoded by a specific group of neurons that are spatially adjacent and produce synchronized spike firing patterns in time. Through this group firing pattern, the symbol "dog" obtains a distributed and robust neuronal representation.

[0034] The core of the symbol encoding module is to use neuron group encoding to represent symbol concepts. Specifically, each symbol is encoded and represented by a specific neuron group, and different neuron groups correspond to different symbols.

[0035] Step S203, using a spiking neural network to process the symbol pulse time sequence, activate the synaptic connections between neuron groups, output the discharge timing of each neuron group, establish logical relationships between different symbols, and perform symbolic reasoning.

[0036] In an embodiment of the present invention, the neuron group interaction module processes the symbol pulse time series from the symbol encoding module, captures the logical correlation between symbols, activates the synaptic connection between neuron groups, outputs the discharge timing of each neuron group, establishes the logical relationship between different symbols, and performs symbolic reasoning to generate reasoning results.

[0037] For example, the concept of "dog" can be encoded by a group of A neurons, and the concept of "animal" can be encoded by another group of B neurons. The two neuron groups are connected by synaptic connections. When the input "dog is an animal" is received, the A neuron group will be stimulated to encode the concept of "dog" and the B neuron group will be stimulated to encode the concept of "animal". At the same time, the synaptic connection between the two is also activated, generating a pulse train.

[0038] Step S204, using the STDP learning mechanism to adjust the synaptic connection strength between neuron groups based on the discharge timing of each neuron group, and outputting the pulses generated by the interaction of the neuron groups.

[0039] In the embodiment of the present invention, the STDP (Spike-Time Dependent Plasticity) learning module is responsible for adjusting the synaptic connection strength between neuron groups so that it can correctly reflect the input symbolic logic knowledge. The working principle of the STDP learning module is to strengthen or weaken the synaptic connection strength between neuron groups based on the discharge timing of each neuron group.

[0040] Step S205, performing symbolic logic reasoning based on the pulses generated by the interaction of the neuron group and combining symbolic logic rules to obtain a symbolic logic reasoning result.

[0041] In the embodiment of the present invention, the symbolic logic reasoning module performs precise logic operations and reasoning according to the pulse output of the neuron group interaction module in combination with the existing symbolic logic rules to obtain the final symbolic logic reasoning result.

[0042] The symbolic reasoning method based on the fusion of symbolic logic and brain-like neural network provided in this embodiment encodes the collected environmental information through the neuron group, maps it into the pulse time series of the neuron group, provides a distributed and robust representation for the symbol, processes the symbol pulse time series using the pulse neural network, provides a brain-like mechanism for symbolic representation and operation, and uses the STDP learning mechanism to simulate the symbolic reasoning process, without the need to explicitly encode complex rules, realizes autonomous learning, combines symbolic logic rules, performs symbolic logic reasoning, and autonomously completes generalized reasoning, which greatly improves the intelligence level and also injects adaptability and robustness into symbolic logic reasoning.

[0043] In this embodiment, a symbolic reasoning method based on the integration of symbolic logic and brain-like neural network is provided, and the process includes the following steps: Step S301, collecting environmental information.

[0044] For details, please see Figure 2 Step S201 of the illustrated embodiment will not be described in detail here.

[0045] Step S302: extract information from the environmental information and represent the extracted information in the form of triples.

[0046] In the embodiment of the present invention, the perception module extracts information from the environmental information and submits the data to the symbol encoding module in the form of a triple (subject, relation, object). The triple is the basis of the knowledge base. The triple is composed of entities (subject, object) and relation words (relation). Taking "dog is an animal" as an example, it can be expressed as dog, is, animal, "dog" and "animal" are entities, and "is" is a conjunction.

[0047] The present invention characterizes the entities and relations in the triples in the form of group neurons, and the existence of group-group synaptic connections between different neuron groups characterizes whether the knowledge represented by different symbol combinations exists and is correct. In actual logical reasoning, these weights also play an important role, enabling the network to complete logical reasoning tasks.

[0048] The synaptic connections between these symbol groups are established through the STDP learning rule. STDP is an important mechanism for simulating the plasticity of biological neurons. It strengthens or weakens the synaptic connection strength between them according to the order of firing time of different neurons. Figure 3 As shown, Figure 3 The workflow of the encoding module is demonstrated, that is, how to encode symbolic logic information and convert it into a pulse sequence that can be processed by SNN. First, the triple knowledge in the structured knowledge base is extracted and integrated into the pulse delay code, which is converted into the electrical stimulation signal received by the neuron group, and finally the discharge at the neuron group level is completed. The discharge of different neuron groups completes the encoding of different symbols.

[0049] By representing the information extracted from the environmental information in the form of triples, it is convenient to send the data in the form of triples for subsequent processing.

[0050] Step S303: using the neuron population code to represent the environmental information, and mapping it into a symbolic pulse time sequence of the neuron population.

[0051] Specifically, the above step S303 includes: Step S3031, assigning neuron groups to entities in the triplet respectively.

[0052] Step S3032, time delay encoding is performed on the relational words in the triplet to obtain a specific time delay between the discharges of the neuron group.

[0053] Step S3033, mapping the specific time delay between the neuron group and the neuron group discharge into the symbol pulse time sequence of the neuron group.

[0054] In an embodiment of the present invention, the symbol encoding module uses a new symbol group encoding algorithm to efficiently encode symbolic logic information. Specifically, the algorithm represents symbolic logic knowledge as a set of triples, each of which corresponds to a minimum logic unit. The algorithm assigns a specific set of neuron populations to each entity in the triple for encoding, and the neuron populations of different entities are connected through synapses. Different entities correspond to different neuron populations, realizing the distributed neural representation of symbols.

[0055] For example, the concept of "dog" can be encoded by a group of A neurons, and the concept of "animal" can be encoded by another group of B neurons. The two groups are connected by synapses, and the strength of the connection indicates the strength of the logical association between "dog" and "dog".

[0056] When the triple "dog is an animal" is input, the A neuron group and the B neuron group will be stimulated successively to generate a pulse sequence. Through the STDP learning rule, the synaptic connection between the two will be strengthened, reflecting the relationship that "dog" belongs to "animal".

[0057] For the encoding of relation words, the algorithm uses time delay encoding. The relation word "is" can be encoded as a specific time delay pattern between the discharges of neuron groups A and B. Through this group encoding method, any complex symbolic logic knowledge can be mapped into the pulse time sequence of the neuron group, laying the foundation for subsequent neuron interaction and reasoning.

[0058] The symbol encoding module maps specific time delays between neuronal groups and neuronal group discharges into symbolic pulse time sequences of the neuronal groups, which are suitable for SNN processing.

[0059] By assigning neuron groups to entities in triples, encoding relational words in triples, and using neuron groups for symbol encoding, a distributed neural representation of symbols is achieved.

[0060] Step S304, using a spiking neural network to process the symbol pulse time sequence, activate the synaptic connections between neuron groups, output the discharge timing of each neuron group, establish the logical relationship between different symbols, and perform symbolic reasoning.

[0061] Specifically, the above step S304 includes: Step S3041, using the bottom layer neurons to encode the single symbol in the symbol pulse time sequence and activate the synaptic connection between the neuron groups.

[0062] Step S3042, using the middle layer neurons to establish the logical relationship between different symbols in the symbol pulse time sequence according to the specific time delay between the discharges of the neuron group.

[0063] Step S3043, using the top layer neurons to perform symbolic reasoning based on the logical relationship between different symbols and combined with context information.

[0064] In the embodiment of the present invention, the neuron group interaction module adopts a pulse neural network, and the neuron group interaction module is composed of multiple layers of SNN. Specifically, the neurons of the neuron group interaction module are organized in layers, and the pulse neural network includes bottom-layer neurons, middle-layer neurons, and top-layer neurons. The bottom-layer neurons encode the concept of a single symbol, the middle-layer neurons capture the logical relationship between different symbols, and the top-layer neurons integrate multiple factors to produce the final reasoning decision output.

[0065] Taking "dog is an animal" as an example, the bottom-level A neuron group will be stimulated to encode the concept of "dog", and the B neuron group will be stimulated to encode the concept of "animal", while activating the synaptic connection between the two. The middle-level neurons can learn the pattern of the discharge timing between the A and B groups, and model the "is-a" logical relationship between "dog" and "animal". Based on the symbolic association modeled by the middle-level neurons, combined with other contextual information (such as context, common sense knowledge, etc.), the top-level decision neurons can correctly output the discharge pattern of "inference result: dog is an animal".

[0066] Similarly, for other complex multi-element combination logical relationships, the neuron group interaction module can also perform high-level symbolic relationship modeling and reasoning judgment through layer-by-layer progressive information transmission and induction.

[0067] The pulse time series is processed by the bottom-layer neurons, middle-layer neurons, and top-layer neurons in the pulse neural network, the logical relationship between symbols is captured, and symbolic reasoning is performed, which injects adaptability and robustness into symbolic reasoning.

[0068] Step S305, using the STDP learning mechanism to adjust the synaptic connection strength between neuron groups based on the discharge timing of each neuron group, and outputting the pulses generated by the interaction of the neuron groups.

[0069] Specifically, the above step S305 includes: Step S3051: if the order in which the neuron groups generate discharges is consistent with the discharge timing, the synaptic connection strength between the neuron groups is strengthened.

[0070] Step S3052: if the order in which the neuron groups generate discharges is inconsistent with the discharge timing, then the synaptic connection strength between the neuron groups is weakened.

[0071] In the embodiment of the present invention, the STDP learning module is responsible for adjusting the connection strength between neuron groups so that they can correctly reflect the input symbolic logic knowledge.

[0072] The working principle of the STDP learning module is that when the A neuron group is excited to generate a pulse train at time t, and the connected B neuron group also discharges after t+Δt, the synaptic connection between A→B will be strengthened. Conversely, if the B neuron group discharges before the A neuron group, the connection between A→B will be weakened.

[0073] This connection adjustment rule based on discharge timing can effectively learn important temporal logical relationships such as "cause and effect" in training samples. For example, by learning "dog is an animal" (A discharges first) and "animal is a living thing" (B discharges first), the system can autonomously establish a topological structure in which A→B synapses are strengthened and B→A synapses are weakened, and correctly map the knowledge hierarchy relationship.

[0074] The STDP module is not only applicable to the initial learning process of symbolic knowledge, but also continuously adjusts the connection of neuron groups in subsequent reasoning and decision-making, so that the system can adapt to new environmental information and task requirements. When there is a deviation between the prediction and the actual reasoning results, STDP will make incremental corrections to the connection between container groups based on the difference in discharge timing, continuously improving the system's reasoning and judgment capabilities.

[0075] By adjusting the synaptic connection strength between neuron groups, the neuron groups can correctly reflect the input symbolic logic knowledge.

[0076] Step S306, performing symbolic logic reasoning based on the pulses generated by the interaction of the neuron groups in combination with symbolic logic rules to obtain symbolic logic reasoning results.

[0077] Specifically, the above step S306 includes: Step S3061, decoding the pulses generated by the interaction of the neuron group to obtain the symbol concept.

[0078] Step S3062, using the pre-established semantic knowledge base to perform symbolic logic reasoning based on symbolic concepts.

[0079] In the embodiment of the present invention, the symbolic logic reasoning module decodes the pulses output by the neuron group interaction module, combines the existing symbolic logic rules, performs precise logic operations and reasoning, and obtains the final symbolic logic reasoning result.

[0080] The symbolic logic reasoning module relies on a pre-established semantic knowledge base, which stores symbolic logic rules such as "reflexivity", "symmetry", "transitivity", etc. These logical axioms lay the theoretical foundation for subsequent symbolic reasoning. When a neuron population outputs certain discharge patterns, the symbolic logic reasoning module will first decode them and determine the corresponding symbolic concept. Then, based on the relationship between concepts and combined with the logical rules in the knowledge base, a series of new symbolic assertions are derived.

[0081] For example, by decoding the two facts "dogs are animals" and "animals are organisms" from the output of the neuron group, the symbolic logic reasoning module can deduce the new conclusion that "dogs are organisms" through the "transitivity" rule. In complex situations, the symbolic logic reasoning module needs to repeatedly apply multiple logic rules, and orderly generate and verify new intermediate assertions according to certain strategies until the required final reasoning result is obtained. The symbolic logic reasoning module provides standardized and interpretable symbolic outputs for the entire system, and can give a clear and understandable logical explanation of the "black box" reasoning process of SNN, further improving the reliability and transparency of the system.

[0082] Specifically, during the learning phase, the system will repeatedly input sample sequences containing symbolic relationships, such as "a dog is an animal." The corresponding neuron groups will be activated and discharge, and the STDP learning module will adjust the weights of the synaptic connections between them according to the timing of the discharges of different groups. After iterative learning of a large number of samples, the system can establish a symbolic relationship knowledge network that reflects rich logical knowledge such as "genus-species relationship" and "cause-effect relationship."

[0083] Once this symbolic neural representation and logical connection are learned, the neuron population interaction module can efficiently perform symbolic reasoning. Given a set of initial symbols as input, the corresponding neuron population will be activated. Then, through synaptic transmission, the activation will gradually spread along the established synaptic connections to other populations connected to it, simulating the process of symbolic reasoning.

[0084] In this process, the interaction and activation patterns between different symbolic groups change dynamically over time. Some groups will be continuously activated, and the corresponding symbolic concepts are the intermediate results of reasoning; while other groups will be inhibited. This interactive process based on neural dynamics can naturally simulate and execute symbolic reasoning rules without explicitly encoding complex logical rules.

[0085] The final result of the reasoning process will be decoded by the symbolic logic reasoning module. This module performs precise symbolic reasoning and calculations based on the activation state of the neuron group output by the system and combines the existing symbolic logic knowledge to output the final symbolic logic reasoning result.

[0086] like Figure 4 As shown, Figure 4 This is a schematic diagram of the symbolic logic reasoning process. In terms of architecture, symbolic logic rules and pulse neural networks are integrated to improve the modeling ability of symbolic logic reasoning. When it is necessary to reason about an external assertion to be reasoned, the assertion to be reasoned will first be converted into a corresponding electrical stimulus through the neuron community encoder, and the corresponding neuron group will be stimulated. Then, the pulse neural network will complete the sequential activation of different neuron groups. Since the network has learned different triplet knowledge and has mastered the knowledge in the form of a neuron community topology, it can perform symbolic logic interpretable reasoning when facing the assertion to be reasoned. Finally, the reasoning result is decoded and output by the decoder.

[0087] The symbolic reasoning method based on the fusion of symbolic logic and brain-like neural networks provided in this embodiment performs precise logical operations and reasoning by combining the pulse outputs generated by the interaction of neuron groups with existing symbolic logic rules, thereby effectively overcoming the lack of interpretability of deep neural networks, improving interpretability and transparency, and facilitating improving the efficiency of human-computer collaboration.

[0088] The core of the present invention is to use the neuron group coding method to represent symbols, and to use the interaction between neuron groups to complete symbolic reasoning. Specifically, the system first establishes synaptic connections between neuron groups through learning and using the STDP principle. Different neuron group codes represent different symbolic concepts, and the connection between groups reflects the logical relationship between symbols.

[0089] Once learning establishes this symbolic neural representation and logical connection, the system can perform symbolic reasoning efficiently. Given a set of initial symbols as input, the corresponding neuronal population will be activated. Then through synaptic transmission, the activation will gradually spread to other connected populations, simulating the process of symbolic reasoning. Eventually, some neuronal populations will be continuously activated, and their corresponding symbolic concepts are the results of reasoning. The advantage of this method is that the neuronal population encoding provides a distributed and robust representation for symbols, while the dynamic interactions between neurons can naturally simulate and execute symbolic reasoning rules. Unlike traditional symbolic systems, this symbolic reasoning based on neural principles does not require explicit encoding of complex logical rules. It only needs to learn symbolic associations through sample training to independently complete generalized reasoning.

[0090] On the other hand, symbolic logic provides SNN with the ability of semantic understanding and abstract reasoning, overcoming the defects of neural network being a "black box" and difficult to explain. By integrating the pattern recognition ability of SNN with symbolic logic, the present invention achieves a deep fusion of perception and reasoning, enabling the intelligent system to simultaneously obtain semantic understanding and abstract reasoning capabilities of the environment, greatly improving the intelligence level of the system.

[0091] The technical solution of the present invention has the following significant advantages: 1. It integrates the advantages of the two major artificial intelligence paradigms of symbolic reasoning and neural networks, breaks through the bottleneck of a single method, and achieves a high degree of unity between precise reasoning and learning generalization. By mapping symbolic concepts into distributed neural representations through neuron group encoding and using neuron dynamics to simulate the symbolic reasoning process, this solution can autonomously learn and perform symbolic logic operations without explicitly encoding complex rules. At the same time, the powerful pattern recognition and generalization capabilities of neural networks also inject adaptability and robustness into symbolic reasoning. The integration of the advantages of the two is expected to give birth to a new generation of general artificial intelligence systems with both reasoning and cognitive capabilities.

[0092] 2. Based on the biological computing principle of pulse neural network, it provides a brain-like mechanism for symbolic representation and operation. Synaptic plasticity learning rules such as STDP simulate the way biological nervous systems acquire knowledge, and can efficiently learn rich symbolic logical relationship knowledge from limited samples. The neuron population encoding and dynamic interaction process also draws on the biological principles of the brain's conceptual representation and reasoning, and has good biological interpretability. Unlike traditional formal symbolic systems, this method is rooted in the origin of biological intelligence and helps to study the neural basis of symbolic reasoning.

[0093] 3. The symbolic logic reasoning module gives a clear and interpretable symbolic output to the "black box" judgment process of SNN. According to the activation state of the SNN neuron group, this module applies the existing logical reasoning rules and semantic knowledge to give a symbolic reasoning path for the judgment made by the system. This mechanism effectively overcomes the lack of interpretability of deep neural networks, improves the interpretability and transparency of the system, and is conducive to improving the efficiency of human-computer collaboration.

[0094] 4. It has good scalability and flexibility, and can adapt to challenging application requirements in different fields. By adjusting the neuron population encoding scheme, STDP learning strategy and symbolic logic reasoning rule base, the architecture can be generalized to different symbolic reasoning tasks, such as machine reading comprehension, visual common sense reasoning, program semantic analysis, etc. At the same time, the system has low requirements for the underlying sensor data, can adapt to multi-modal heterogeneous input, and provide a general perception and reasoning engine for various artificial intelligence products and services.

[0095] Specifically, for the symbol encoding module in this embodiment, in addition to the symbol group encoding algorithm, other encoding methods such as vector encoding, tensor encoding, etc. can also be considered to adapt to different symbolic logic information.

[0096] For the STDP learning module in this embodiment, in addition to the STDP learning algorithm, other learning rules such as reinforcement learning, unsupervised learning, etc. can also be considered to improve learning efficiency and reasoning accuracy.

[0097] For the symbolic logic reasoning module in this embodiment, in addition to the symbolic logic reasoning module, other reasoning methods, such as fuzzy logic, probabilistic reasoning, etc., can also be considered to meet different reasoning requirements.

[0098] In this embodiment, a symbolic reasoning system based on the fusion of symbolic logic and brain-like neural network is provided, which is used to implement the above-mentioned embodiments and preferred implementation modes, and will not be repeated hereafter. As used below, the term "module" can be a combination of software and / or hardware that implements a predetermined function. Although the system described in the following embodiments is preferably implemented in software, the implementation of hardware, or a combination of software and hardware, is also possible and conceivable.

[0099] This embodiment provides a symbolic reasoning system based on the integration of symbolic logic and brain-like neural network. Figure 5 As shown, the system includes: The perception module 501 is used to collect environmental information.

[0100] The symbol encoding module 502 is used to use neuron group encoding to represent environmental information, and map it into a symbol pulse time sequence of the neuron group. Different neuron groups are connected through synapses, and neurons generate synchronous pulse discharge patterns.

[0101] The neuron group interaction module 503 is used to process the symbol pulse time series using the pulse neural network, activate the synaptic connection between the neuron groups, output the discharge timing of each neuron group, establish the logical relationship between different symbols, and perform symbolic reasoning.

[0102] The STDP learning module 504 is used to adjust the synaptic connection strength between neuron groups based on the discharge timing of each neuron group using the STDP learning mechanism, and output the pulses generated by the interaction of the neuron groups.

[0103] The symbolic logic reasoning module 505 is used to perform symbolic logic reasoning based on the pulses generated by the interaction of the neuron group and in combination with the symbolic logic rules to obtain the symbolic logic reasoning results.

[0104] Specifically, the perception module extracts information from the environment and passes the data to the symbolic encoding module in the form of triplets. The symbolic encoding module encodes the triplet information as electrical stimulation of the neuron group and the topological structure corresponding to the knowledge is learned by the spiking neural network through the STDP learning rule. In the process of symbolic logic reasoning, the topological structure of the synaptic connections between different neuron groups in the network ensures the accuracy and interpretability of logical reasoning.

[0105] In some optional embodiments, the system further comprises: The extraction module is used to extract information from environmental information and represent the extracted information in the form of triples, which are composed of entities and relation words.

[0106] In some optional implementations, the symbol encoding module 502 includes: The first encoding unit is used to assign neuron groups to entities in the triplet respectively, wherein the neuron groups of different entities are connected through synapses.

[0107] The second encoding unit is used to perform time delay encoding on the relational words in the triplet to obtain a specific time delay between the discharges of the neuron group.

[0108] A mapping unit is used to map a specific time delay between a neuron group and a neuron group discharge into a symbolic pulse time sequence of the neuron group, and the symbolic pulse time sequence is suitable for SNN processing.

[0109] In some optional embodiments, the neuron population interaction module 503 includes: The third encoding unit is used to encode a single symbol in the symbol pulse time sequence using the bottom layer neurons and activate the synaptic connection between the neuron groups.

[0110] The logic relationship establishing unit is used to establish the logic relationship between different symbols in the symbol pulse time sequence by using the middle layer neurons according to the specific time delay between the discharges of the neuron group.

[0111] The symbolic reasoning unit is used to perform symbolic reasoning based on the logical relationship between different symbols and context information using the top-layer neurons.

[0112] In some optional implementations, the STDP learning module 504 includes: The first adjustment unit is used to strengthen the synaptic connection strength between the neuron groups if the order in which the neuron groups generate discharges is consistent with the discharge timing.

[0113] The second adjustment unit is used to weaken the synaptic connection strength between the neuron groups if the order in which the neuron groups generate discharges is inconsistent with the discharge timing.

[0114] In some optional implementations, the symbolic logic reasoning module 505 includes: The decoding unit is used to decode the pulses generated by the interaction of the neuron group to obtain the symbolic concept.

[0115] The logic reasoning unit is used to perform symbolic logic reasoning according to symbolic concepts using a pre-established semantic knowledge base, wherein the pre-established semantic knowledge base stores symbolic logic rules.

[0116] The further functional description of each of the above modules and units is the same as that of the above corresponding embodiments and will not be repeated here.

[0117] The symbolic reasoning system based on the fusion of symbolic logic and brain-like neural network in this embodiment is presented in the form of functional units, where the units refer to ASIC (Application Specific Integrated Circuit) circuits, processors and memories that execute one or more software or fixed programs, and / or other devices that can provide the above functions.

[0118] The embodiment of the present invention also provides a computer device having the above Figure 5 The symbolic reasoning system shown is based on the fusion of symbolic logic and brain-like neural networks.

[0119] See also Figure 6 , Figure 6 is a schematic diagram of the structure of a computer device provided by an optional embodiment of the present invention, such as Figure 6 As shown, the computer device includes: one or more processors 10, a memory 20, and interfaces for connecting various components, including high-speed interfaces and low-speed interfaces. Various components are connected to each other using different buses for communication, and can be installed on a common mainboard or installed in other ways as needed. The processor can process the instructions executed in the computer device, including instructions stored in or on the memory to display the graphical information of the GUI on an external input / output device (such as, a display device coupled to the interface). In some optional embodiments, if necessary, multiple processors and / or multiple buses can be used together with multiple memories and multiple memories. Similarly, multiple computer devices can be connected, and each device provides some necessary operations (for example, as a server array, a group of blade servers, or a multi-processor system). Figure 6 A processor 10 is taken as an example.

[0120] The processor 10 may be a central processing unit, a network processor or a combination thereof. The processor 10 may further include a hardware chip. The hardware chip may be a dedicated integrated circuit, a programmable logic device or a combination thereof. The programmable logic device may be a complex programmable logic device, a field programmable gate array, a general purpose array logic or any combination thereof.

[0121] The memory 20 stores instructions executable by at least one processor 10, so that at least one processor 10 executes the method shown in the above embodiment.

[0122] The memory 20 may include a program storage area and a data storage area, wherein the program storage area may store an operating system, an application required for at least one function; the data storage area may store data created according to the use of the computer device, etc. In addition, the memory 20 may include a high-speed random access memory, and may also include a non-transient memory, such as at least one disk storage device, a flash memory device, or other non-transient solid-state storage device. In some optional embodiments, the memory 20 may optionally include a memory remotely arranged relative to the processor 10, and these remote memories may be connected to the computer device via a network. Examples of the above-mentioned network include, but are not limited to, the Internet, an intranet, a local area network, a mobile communication network, and combinations thereof.

[0123] The memory 20 may include a volatile memory, such as a random access memory; the memory may also include a non-volatile memory, such as a flash memory, a hard disk or a solid state drive; the memory 20 may also include a combination of the above types of memory.

[0124] The computer device also includes an input device 30 and an output device 40. The processor 10, the memory 20, the input device 30 and the output device 40 may be connected via a bus or other means. Figure 6 The example of connecting through bus is taken in the following.

[0125] The input device 30 can receive input digital or character information and generate key signal input related to user settings and function control of the computer device, such as a touch screen, etc. The output device 40 can include a display device, etc.

[0126] The embodiment of the present invention also provides a computer-readable storage medium. The method according to the embodiment of the present invention can be implemented in hardware, firmware, or can be implemented as a computer code that can be recorded in a storage medium, or can be implemented as a computer code that is originally stored in a remote storage medium or a non-temporary machine-readable storage medium and will be stored in a local storage medium through a network download, so that the method described herein can be stored in such software processing on a storage medium using a general-purpose computer, a dedicated processor, or programmable or dedicated hardware. Among them, the storage medium can be a magnetic disk, an optical disk, a read-only storage memory, a random access memory, a flash memory, a hard disk or a solid-state hard disk, etc.; further, the storage medium can also include a combination of the above types of memories. It can be understood that a computer, a processor, a microprocessor controller, or programmable hardware includes a storage component that can store or receive software or computer code. When the software or computer code is accessed and executed by a computer, a processor, or hardware, the method shown in the above embodiment is implemented.

[0127] A part of the present invention may be applied as a computer program product, such as a computer program instruction, which, when executed by a computer, can call or provide the method and / or technical solution according to the present invention through the operation of the computer. Those skilled in the art should understand that the existence of the computer program instruction in a computer-readable medium includes, but is not limited to, a source file, an executable file, an installation package file, etc., and accordingly, the way in which the computer program instruction is executed by the computer includes, but is not limited to: the computer directly executes the instruction, or the computer compiles the instruction and then executes the corresponding compiled program, or the computer reads and executes the instruction, or the computer reads and installs the instruction and then executes the corresponding installed program. Here, the computer-readable medium may be any available computer-readable storage medium or communication medium accessible to the computer.

[0128] Although the embodiments of the present invention have been described in conjunction with the accompanying drawings, those skilled in the art may make various modifications and variations without departing from the spirit and scope of the present invention, and such modifications and variations all fall within the scope of the present application.

Claims

1. A symbolic reasoning method based on the integration of symbolic logic and brain-like neural network, characterized in that: The method comprises: Collect environmental information; The environmental information is represented by coding a neuron population and mapped into a symbolic pulse time sequence of the neuron population. Different neuron populations are connected via synapses, and the neurons generate a synchronous pulse discharge pattern; Processing the symbol pulse time sequence using a spiking neural network, activating synaptic connections between neuron groups, outputting the discharge timing of each neuron group, establishing logical relationships between different symbols, and performing symbolic reasoning; Using the STDP learning mechanism, the synaptic connection strength between the neuron groups is adjusted based on the discharge timing of each neuron group, and the pulses generated by the interaction of the neuron groups are output; According to the pulses generated by the interaction of the neuron groups, symbolic logic reasoning is performed in combination with symbolic logic rules to obtain symbolic logic reasoning results.

2. The method according to claim 1, characterized in that After collecting the environmental information, the method further includes: Information is extracted from the environmental information, and the extracted information is represented in the form of triples, where the triples are composed of entities and relational words.

3. The method according to claim 2, characterized in that The method of using a neuron population to encode and represent the environmental information and mapping it into a symbol pulse time sequence of the neuron population includes: Assign neuron populations to entities in the triplet respectively, wherein neuron populations of different entities are connected through synapses; The time delay encoding of the relational words in the triplet results in a specific time delay between the firing of the neuronal population; The neuron population and specific time delays between the neuron population discharges are mapped to a symbolic pulse time sequence of the neuron population, which is suitable for SNN processing.

4. The method according to claim 3, characterized in that: The spiking neural network includes bottom-layer neurons, middle-layer neurons, and top-layer neurons. The spiking neural network is used to process the symbol pulse time sequence, activate synaptic connections between neuron groups, output the discharge timing of each neuron group, establish logical relationships between different symbols, and perform symbolic reasoning, including: Encoding a single symbol in the symbol pulse time sequence using the bottom layer neurons and activating synaptic connections between neuron populations; Using the middle layer neurons to establish a logical relationship between different symbols in the symbol pulse time sequence according to a specific time delay between the discharges of the neuron population; The top layer of neurons is used to perform symbolic reasoning based on the logical relationship between the different symbols in combination with context information.

5. The method according to claim 1, characterized in that The step of adjusting the synaptic connection strength between neuron groups based on the discharge timing of each neuron group includes: If the order in which the neuron groups generate discharges is consistent with the discharge timing, the synaptic connection strength between the neuron groups is strengthened; If the order in which the neuron groups generate discharges is inconsistent with the discharge timing, the synaptic connection strength between the neuron groups is weakened.

6. The method according to claim 1, characterized in that The step of performing symbolic logic reasoning based on the pulses generated by the interaction of the neuron population and combining symbolic logic rules to obtain a symbolic logic reasoning result includes: Decoding the pulses generated by the interaction of the neuron population to obtain symbolic concepts; Symbolic logic reasoning is performed according to the symbolic concept using a pre-established semantic knowledge base, wherein the pre-established semantic knowledge base stores symbolic logic rules.

7. A symbolic reasoning system based on the integration of symbolic logic and brain-like neural networks, characterized in that: The system comprises: Perception module, used to collect environmental information; A symbol encoding module, used to represent the environmental information by encoding the neuron population, and mapping it into a symbol pulse time sequence of the neuron population. Different neuron populations are connected by synapses, and the neurons generate a synchronous pulse discharge pattern; A neuron population interaction module is used to process the symbol pulse time sequence using a spiking neural network, activate synaptic connections between neuron populations, output the discharge timing of each neuron population, establish logical relationships between different symbols, and perform symbolic reasoning; An STDP learning module, used to adjust the synaptic connection strength between neuron groups based on the discharge timing of each neuron group using the STDP learning mechanism, and output pulses generated by the interaction of the neuron groups; The symbolic logic reasoning module is used to perform symbolic logic reasoning based on the pulses generated by the interaction of the neuron groups in combination with symbolic logic rules to obtain symbolic logic reasoning results.

8. A computer device, characterized in that: include: A memory and a processor, wherein the memory and the processor are communicatively connected to each other, the memory stores computer instructions, and the processor executes the symbolic reasoning method based on the fusion of symbolic logic and brain-like neural network as described in any one of claims 1 to 6 by executing the computer instructions.

9. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores computer instructions, and the computer instructions are used to enable a computer to execute the symbolic reasoning method based on the fusion of symbolic logic and brain-like neural network as described in any one of claims 1 to 6.

10. A computer program product, characterized in that It includes computer instructions, which are used to enable a computer to execute the symbolic reasoning method based on the fusion of symbolic logic and brain-like neural network as described in any one of claims 1 to 6.

Images