66 results about "Thinking processes" patented technology

The invention relates to a model training method and device, equipment, a storage medium and a product. The method comprises the following steps: constructing a training data set according to a target text reasoning chain obtained by converting multi-modal data; according to the training data set, performing supervision fine tuning on the pre-trained multi-modal large language model to obtain a basic reasoning model; performing optimization processing on the basic reasoning model according to reinforcement learning training of long thinking to obtain a target reasoning model; the target reasoning model is used for outputting a target answer containing a reasoning process according to the input multi-modal data. Therefore, the long text constraint can be directly used for reinforcement learning, and the training efficiency is greatly improved; and by adopting long-thinking reinforcement learning training, the model can easily learn a correct thinking process in training, so that the reasoning ability of the multi-modal large language model for processing a complex visual reasoning task is improved, and the correct thinking process is displayed in the reasoning process.

A tree-of-thought (ToT) system is presented that improves problem-solving capabilities of machine learning models, such as auto-regressive large language models (LLMs). The TOT system can solve complex reasoning tasks through trial and error. In this process, the system explores the solution space through a tree-like thought process, allowing for backtracking when necessary. The system augments an LLM with additional modules including a prompter agent, a checker module, a memory module, and a ToT controller. These modules engage in a multi-round conversation with the LLM. The memory module records the conversation and state history of the problem-solving process, which allows the system to backtrack to the previous steps of the thought-process and explore other solution paths. This new system can be applied to a blockchain and / or a distributed computing system.

Disclosed in the present invention is a general-purpose intelligent agent. The general-purpose intelligent agent comprises: an input module, which is configured to acquire a preprocessed input signal; a consciousness module, which is configured to perform perception and cognitive operations in a transparent and interpretable abstract thinking form; a self-awareness module, which is configured for perception and cognition of an intelligent agent body; a subconsciousness module, which is configured to use a data-driven model to implement perception and cognitive functions at a subconscious level; an information exchange module, which is configured to perform mutual exchange between perception, cognition and other information generated by the consciousness module, the self-awareness module and the subconsciousness module, thereby enabling integration and utilization; and an output module, which is configured to output a processing result on demand. In a machine intelligent agent of the present invention, a machine uses a general-purpose language as an underlying language for thinking and interaction, a thinking process and result are completely transparent, and thus autonomous decision-making, autonomous learning and continuous evolution are realized under fully controllable conditions. Further disclosed is a method for constructing a general-purpose intelligent agent and a general-purpose language. Compared with the prior art in which a natural language is used, the machine intelligent agent based on the general-purpose language has wide versatility in application.

Systems and methods for guiding or steering thought processes on a persistent cognitive machine (PCM) that uses a continuous, differentiable, thought manifold in geometric space to allow a computer to engage in human-like thought processes. The PCM with thought manifold represents a fundamental advancement in artificial intelligence beyond current probabilistic AI system such as large language models (LLMs) and similar reasoning models. A PCM with cognitive manifold performs cognition on a thought manifold in a continuous, differentiable, thought manifold in geometric space as opposed to probabilistic prediction in a discontinuous, anisotropic, and topologically fractured vector space. Methods for guiding or steering thought processes on the thought manifold are disclosed that involve mathematical manipulations of the geometric space of the thought manifold inspired by gravitational lensing.

A system and method for a digital thought architecture, otherwise called a persistent cognitive machine (PCM), that uses a continuous, differentiable, thought manifold in geometric space to allow a computer to engage in human-like thought processes. The PCM with thought manifold represents a fundamental advancement in artificial intelligence beyond current probabilistic AI system such as large language models (LLMs) and similar reasoning models. Not only does the PCM with thought manifold maintain persistent cognitive processes regardless of external interaction, overcoming limitations of existing AI systems that operate within a prompt-response paradigm where they await input, generate output, and return to a waiting state, it also performs cognition on a thought manifold in a continuous, differentiable, thought manifold in geometric space as opposed to probabilistic prediction in a discontinuous, anisotropic, and topologically fractured vector space. In some embodiments, the thought manifold may be implemented as a neuromorphic platform.

The invention belongs to the technical field of electric power artificial intelligence, and discloses an electric power knowledge question-answering method based on a knowledge reasoning thinking chain embedded large model and a related device. The method comprises the steps of obtaining a power problem text; and inputting the power question text into a pre-trained knowledge reasoning thinking chain embedding large model, and outputting a text answer of the power question. According to the method, a supervised learning and preference optimization mechanism is fused, and collaborative optimization of large-model professional knowledge learning and thinking path modeling is realized by constructing an electric power knowledge question and answer and analyzing a thinking chain data set; the technical problems that an existing large model is weak in knowledge, inconsistent in thinking process and insufficient in generalization ability in power professional text questions and answers are solved.

The invention provides a method for realizing agent thinking framework development through computing power of an intelligent computing center. The method comprises the following steps: S1, a thinking framework development agent receives task information for constructing a target agent thinking framework; s2, determining development requirements according to the task information; s3, determining a design scheme according to development requirements; s4, generating a code of a thinking framework and a thinking process output code according to the development requirement and the design scheme; s5, deploying a running environment, debugging codes of the thinking framework and outputting thinking process information; when the debugging result does not meet the development requirement, checking the design scheme and the code, returning to the step S3 if the design scheme needs to be modified, and returning to the step S4 if the code needs to be modified until the development requirement is met; s6, evaluating the thinking process information, if the thinking process information does not pass the evaluation and the design scheme and the codes are checked, returning to the step S3 if the design scheme needs to be modified, and returning to the step S4 if the codes need to be modified until the thinking process information passes the evaluation; and S7, feeding back result information.

A head-mountable device can facilitate user thought processes by recording user-perceivable experiences during a first mode while the head-mountable device is operated in a capture mode. While in the capture mode, the head-mountable device can record inputs from the user. During a second mode, the head-mountable device can reproduce the previously recorded experiences as well as the user inputs so that the user can resume development of the thoughts and ideas associated with the first mode. The head-mountable device can also track the user's conditions to monitor attention levels of the user and provide indicators to prompt a user to perform activities that will help the user refocus and / or provide additional inputs.

The invention provides a corpus data generation method and device based on a large model and an intelligent agent, and relates to the technical field of artificial intelligence, in particular to the technical fields of deep learning, large models, intelligent questions and answers and the like. The corpus data generation method based on the large model comprises the steps that a content generation task is executed on the basis of a preset requirement condition through the large model, corpus content is obtained, the content generation task comprises multiple target tasks with a dependency relationship, and the multiple target tasks represent the thinking process of the large model for the corpus content to be generated; and based on the corpus content and the thinking process information related to the multiple target tasks, determining target corpus data.

The invention discloses a method and a device for improving safety capability of a generative model based on thinking chain reinforcement, and formulates a two-stage training framework in order to reinforce a generation path of a thinking chain. In the first stage, thinking data generated by the model is utilized to carry out supervised fine tuning on a target model to limit an action space of the model in a formatted reasoning space; in the second stage, in order to enhance the reasonability of the reasoning process in the first stage, the reasoning action of the model is doubly scored at the statement level and the token level so as to screen out offline data to further reinforce learning of the target model. The method has the following advantages that the generalization ability of the model in different attack environments is greatly improved by enhancing the thinking ability of the model; compared with a traditional reward model, a double-level scoring mechanism adopted by the method can efficiently evaluate the model thinking process while keeping the fine granularity of reward signals.

The invention belongs to the technical field of artificial intelligence, and particularly relates to a reinforcement learning agent applied to molecular design, a problem evaluator evaluates and preprocesses molecular design related problems input by a user, outputs user requirements and transmits the user requirements to a decision thinking center module; the decision thinking center module outputs a plurality of thinking processes and matched execution actions based on user requirements, and recommends actions to the action execution module; the action execution module executes actions according to the recommendation of the decision thinking center module; when the action recommended by the decision thinking center module is a summary reply, the answer evaluator evaluates whether the reply passes or rejects the answer; and the reinforcement learning module records and samples multiple thinking processes and execution action tracks and operation results to obtain relative advantages, and continuously optimizes and decides parameters of a model adopted by the thinking center module by utilizing the optimal thinking process. The decision-making path of the reinforcement learning agent is simple, the efficiency is high, and continuous learning can be realized.

Systems and methods for latent slice budgeting on a persistent cognitive machine (PCM) that uses a continuous, differentiable, cognitive manifold in geometric space to allow a computer to engage in human-like thought processes. The PCM with cognitive manifold represents a fundamental advancement in artificial intelligence beyond current probabilistic AI system such as large language models (LLMs) and similar reasoning models. A PCM with cognitive manifold performs cognition on a thought manifold in a continuous, differentiable, thought manifold in geometric space as opposed to probabilistic prediction in a discontinuous, anisotropic, and topologically fractured vector space. Methods for latent slice budgeting on the cognitive manifold are disclosed that foliation of the cognitive manifold into time slices and budgeting change between the time slices.

The embodiment of the invention provides a model parameter determination method and device, a storage medium and a program product, and the method comprises the steps: processing input data through employing a specified model, so as to generate at least one piece of thinking process information and corresponding answer information, and at least part of the thinking process information corresponds to a plurality of pieces of answer information; performing evaluation processing on the input data based on at least part of the answer information to determine corresponding optimization feedback information; updating at least one parameter of the specified model according to the optimization feedback information; wherein the specified model is a visual language model or a large language model, and the input data comprises at least one of text data, image data and point cloud data. Through multi-answer sampling, optimization feedback of more answer levels can be obtained under the condition that the generation number in the thinking process is reduced or is kept unchanged, so that the training time and the calculation overhead are remarkably reduced, the training efficiency is improved, and the thinking and summarization ability of the model can be improved under limited calculation resources through a training scheme after reinforcement learning.

The invention belongs to the field of large model training, and particularly relates to a multi-modal data annotation and large model thinking chain training method based on visual interaction. The method comprises the following steps: 1, acquiring multi-modal data; 2, displaying the multi-modal data on a display, displaying the multi-modal data to an expert through the display, recording an eye movement track when the expert watches the multi-modal data on the display by adopting an eye movement data acquisition device, and acquiring voice information of the expert by adopting a voice acquisition device; 3, according to the eye movement track and the voice information, forming multi-modal annotation data; and 4, preprocessing the multi-modal annotation data, and inputting the preprocessed multi-modal annotation data into the large model for training to obtain a trained large model. According to the method, the multi-modal labeling data of the experts in the labeling process can be collected, dominant processing is carried out on the thinking chain and the thinking process of the experts, logic fusion is carried out, and the information breadth and accuracy of a large model are improved.

The application provides a method for generating an answer to a question, an electronic device and a computer program product, and applies to the technical field of artificial intelligence. The method comprises the following steps: in response to an input question to be replied, a plurality of reply rules representing a thinking process in an auxiliary learning scenario are determined; the reply rules are rules that need to be followed when replying to the question; the plurality of reply rules are sorted differently, different thinking processes are simulated based on different sorting results, and the reply content generated under the optimal thinking process is selected as the answer to the question to be replied; wherein the reply content generated under the thinking process is generated based on the reply content corresponding to each reply rule included in the thinking process. The application designs a plurality of reply rules representing a thinking process in an auxiliary learning scenario, and combines different reply rule sequences, so that the reply process can reflect the human thinking process. Compared with directly generating an answer based on model parameters, the intelligent level is improved.

The invention relates to a brain-like scene safety assessment method combining process supervision and fast and slow thinking. The method comprises the following steps: firstly, designing a brain-like scene security cognition process and a scene security level division standard; then, constructing a supervision fine tuning data set and a reinforcement learning fine tuning data set of a process supervision normal form based on the brain-like scene security cognition process and a scene security level division standard; then, performing cold start on an open-source pre-training multi-modal large language model based on the supervision fine tuning data set to obtain a supervision fine tuning model; carrying out reinforcement learning training on the supervision fine tuning model based on the reinforcement learning fine tuning data set of the process supervision normal form; and finally, inputting a to-be-evaluated scene image into the trained supervision fine tuning model to obtain a scene security level. The inference process of the model can be aligned with the thinking process of human in the macroscopic level, the flexibility of the thinking process is maintained, and the interpretability of the result of each step is improved.

The present invention relates to the technical field of artificial intelligence, and provides a method and apparatus for constructing a human-like answering engine. The method comprises: acquiring a state transition graph of each answering algorithm; respectively merging and integrating input nodes, output nodes, identical key state nodes, and identical state transition paths in all state transition graphs to form an integrated state transition graph; and performing function configuration on the integrated state transition graph to construct a static graph of an answering technique as the human-like answering engine. The method and apparatus for constructing the human-like answering engine provided by the present invention simulate the thought processes of human, and progressively analyze and solve algebraic questions by means of a series of state transitions, so that the answering engine is more flexible and intelligent, and can handle complex and variable question types, thereby expanding the range of answerable questions.

The invention discloses a fine tuning method, system and equipment for an electric power intrinsic safety analysis model based on knowledge distillation, and a medium, and belongs to the technical field of electric power safety analysis. The method comprises the following steps: collecting professional knowledge data in the field of electric power intrinsic safety analysis, and performing professional knowledge supervision fine tuning to obtain a distillation model; and guiding an output thinking process and a root cause based on problem description through the distillation model, and introducing a super model and a causal knowledge graph for evaluation to obtain an evaluation result. Through combination of knowledge distillation and the causal knowledge graph, the accuracy and interpretability of an electric power intrinsic safety analysis inference model are remarkably improved, the thinking process and the root cause of problem description are generated through the distillation model, output is evaluated by using the super model and the causal knowledge graph, the causal inference ability is enhanced, and the accuracy and the interpretability of the electric power intrinsic safety analysis inference model are improved. The problem that the causal relationship of a traditional model is fuzzy in complex power accident analysis is effectively solved.

Systems and methods for persistence of memory on a persistent cognitive machine (PCM) that uses a continuous, differentiable, cognitive manifold in geometric space to allow a computer to engage in human-like thought processes. The PCM with cognitive manifold represents a fundamental advancement in artificial intelligence beyond current probabilistic AI system such as large language models (LLMs) and similar reasoning models. A PCM with cognitive manifold performs cognition on a thought manifold in a continuous, differentiable, thought manifold in geometric space as opposed to probabilistic prediction in a discontinuous, anisotropic, and topologically fractured vector space. Persistence of memory is reflected on the cognitive manifold through relative displacements between geodesics after a reasoning trajectory has been calculated in a manner analogous to gravitational wave echoes in general relativity physics.

The invention relates to a visual thinking chain, in particular to a bionic visual thinking chain cross-modal model based on eye movement tracking and a visualization system and method, and solves the technical problems that an existing visual thinking chain model lacks real visual thinking process simulation and is difficult to establish a visual thinking chain conforming to a real human visual thinking mode. According to the bionic visual thinking chain cross-modal model based on eye movement tracking, the to-be-detected image eye movement gazing area image blocks from eye movement tracking data and the gazing sequence and the gazing duration of the to-be-detected image eye movement gazing area image blocks are adopted, and the dynamic characteristics and the time sequence mode of human visual cognition can be truly reflected; the big language model is made to process visual information step by step according to the sequence of human eye movement tracks, a human real visual thinking mode is simulated, and therefore a real visual thinking chain is established; besides, an eye movement vector prediction matrix and an eye movement prediction text answer can be obtained, so that a bionic visual thinking chain and a text thinking chain are output at the same time, and the interpretability of the thinking process after the text and the image are fused is improved.

The invention discloses a Chinese grammar error correction multi-task training method based on thinking process distillation, which comprises the following steps: in a data construction stage, guiding a DeepSeek model by utilizing a designed constraint condition and a prompt project, and taking the DeepSeek model as a teacher model, in a training stage, taking a small-volume large language model as a student model to learn knowledge of the teacher model, and taking a small-volume large language model as a student model to learn the knowledge of the teacher model. The semantic comprehension ability of a teacher model is migrated from multiple dimensions by combining structured supervision and a multi-task learning mechanism, so that tasks are mutually enhanced, the same wrong sentence can be trained at the same time in multiple dimensions, and the model is promoted to learn more universal and more robust feature representation; according to the method, an end-to-end interpretable error correction process is realized, a user is helped to understand a model decision basis, and an error correction analysis and display process is added; the features of different dimensions are aligned in the multi-task learning process, so that the model learns more universal and more robust feature representation, and the generalization ability of the model in complex grammar error, composite error and zero sample scenes is remarkably improved.

The application discloses a trial finding text generation method and related device, and relates to the technical field of artificial intelligence, which comprises the following steps: normalizing a plurality of evidence data of a target case into a structured evidence list, taking an evidence item in the evidence list as an evidence node, taking a fact element extracted from the evidence list as a fact node, constructing an evidence fact correlation graph, the edge relationship in the evidence fact correlation graph comprises a contradiction relationship, inputting the evidence fact correlation graph into a graph neural network model for feature transmission and aggregation between nodes, obtaining updated features of each fact node, and determining a dispute focus node and a non-dispute node from each fact node according to the fact credibility in the updated features and the contradiction relationship in the evidence fact correlation graph, and generating a trial finding text in a judgment according to the dispute focus node and the non-dispute node. The application simulates the reasoning thinking process of judges in determining facts by the graph node feature transmission process, and improves the accuracy and efficiency of fact determination.

The invention discloses an interpretable sleep staging method based on training after visual language model supervision fine tuning, which comprises the following steps: acquiring multi-channel sleep data and a corresponding sleep reasoning text to construct a data set; performing numbering based on interpretation rules related to sleep in clinical rule knowledge to construct a corresponding rule base; taking a pre-trained visual language model as a basic model, and performing supervision fine tuning training on the basic model through the data set to obtain a visual language model; multi-channel sleep data to be analyzed and preset cue words are input into the visual language model to generate a structured JSON result, and the JSON result comprises a sleep stage label, a judgment reason text and a rule number list of related interpretation rules. The invention further provides a device capable of explaining sleep staging. According to the method provided by the invention, the thinking process of experts can be simulated, so that comprehensive data support conforming to clinical rules is provided for each sleep staging result.

The invention discloses a medical history collection system supporting clinical thinking path recognition, and belongs to the technical field of medical information. The system comprises a hypothesis agent cluster management module used for initializing an agent cluster composed of a plurality of hypothesis agents based on a clinical knowledge graph, each hypothesis agent corresponding to a candidate disease and comprising static knowledge related to the candidate disease; the doctor-patient interaction interface module is used for receiving the initial clinical evidence and analyzing the initial clinical evidence into an initial clinical evidence state vector; the dynamic evidence processing and state evolution engine is used for iteratively driving the state evolution of the agent cluster; and the thinking path generation and visualization module is used for generating a clinical thinking path according to the state evolution process of the agent cluster. The technical problem that the dynamic and nonlinear diagnosis thinking process of clinicians cannot be effectively captured and restored in the prior art is solved.

The invention relates to the technical field of operation and maintenance of communication operators, in particular to a method and system for complementing a positioning rule based on large model reasoning ability, and the method comprises the following steps: fault feature and tool preparation, cue word framework construction, thinking process generation and expert auditing, rule generation and current network verification, and feedback optimization mechanism. The method has the beneficial effects that by virtue of the reasoning capability of the Deepseek large model, the rule data of fault diagnosis is generated by continuously adjusting the input conditions such as the fault feature library and the fault diagnosis tool. Supplementing or perfecting a current network fault positioning rule; the problems of high expert dependency, insufficient rule credibility, difficulty in verification and the like in a traditional rule are solved, the time investment of experts is reduced, and the existing rule is perfected; an efficient, credible and iterable fault positioning mechanism is realized, and the intelligent level of network operation and maintenance is improved.

A method for controlling an artificial intelligence (AI) device can include receiving a user query, retrieving safe and unsafe trajectories from a trajectory history database, providing the user query, the at least one safe trajectory, and the at least one unsafe trajectory to an actor agent configured as a first large language model, generating, by the actor agent, a proposed action and a thought process for performing a step related to the task based on the at least one safe trajectory and the at least one unsafe trajectory. Also, the method can further include providing the proposed action and the thought process to a critic agent configured as a second large language model, generating, by the critic agent, a feedback critique of the proposed action, and executing, by the actor agent, a final action in an environment based on the feedback critique from the second large language model.

The invention discloses a visual shared whiteboard system and method for solving a group collaboration mathematical problem, and belongs to the technical field of computer science and human-computer interaction crossing. The system comprises a multi-modal fusion input module used for converting input such as handwriting and voice into a structured mathematical object; the real-time semantic understanding and knowledge graph engine is used for constructing a collaborative cognitive model; the collaborative process management module based on the directed hypergraph is used for carrying out formalized modeling and navigation on a nonlinear problem solving process; the context-aware intelligent auxiliary module is used for providing verification, recommendation and guiding consensus achievement; and a multi-user cooperation engine ensures stable cooperation experience. According to the method, the problems of professional expression, real-time cooperation splitting, group thinking process management difficulty and the like are solved, and a unified, intelligent and efficient integrated cooperation environment is provided for group solving of complex mathematical problems.

The invention discloses a micro-course extraction method, a related method and a related device, and relates to the technical field of resource fragmentization, and the micro-course extraction method comprises the steps: obtaining a target course video of a target subject; analyzing the content of the target course video, and generating target text data with a timestamp; performing knowledge point slicing on the target text data based on a large slicing model to obtain a segmentation result at least comprising starting and ending time of a knowledge point fragment, and adopting training text data marked with a real knowledge point segmentation result by the large slicing model, and meanwhile, obtaining thinking process data of the real knowledge point segmentation result in an assisted manner to obtain a knowledge point segmentation result; training the general large model base to obtain training text data, wherein the training text data is text data corresponding to course video samples of the target subject; and extracting the knowledge point video clip from the target course video according to the starting and ending time of the knowledge point clip to obtain a knowledge point video micro-course. The micro-course extraction method disclosed by the invention can automatically extract the accurate knowledge point video micro-course from the course video.

This invention relates to the field of financial technology and discloses a business processing method, electronic device, storage medium, and program product. The method includes: acquiring a user's business requirement description information and parsing the information to obtain the user's business processing requirements under a target business type; supplementing the business processing requirements with requirement description data based on prompts and / or thought processes corresponding to the target business type to obtain business processing association parameters; orchestrating the business processing flow using a trained orchestration model based on the business processing requirements and association parameters to generate a dynamic service plugin invocation scheme; and invoking the corresponding service plugin to perform business processing under the target business type based on the dynamic service plugin invocation scheme and association parameters. By supplementing the business processing requirements with data, accurate matching of requirements can be achieved; and by orchestrating the business processing flow, multiple types of businesses can be processed using a large model.

The invention provides an information determination method, equipment and a computer program product. The method comprises the following steps: determining a text representation of a target hypertext markup language corresponding to a multi-mode user interface of screen information of virtualized mobile equipment; wherein the screen information is generated based on input of a user on the virtualized mobile device; based on the text representation and a large language model, generating a thinking chain prompt and a to-be-executed task corresponding to the text representation; wherein the thinking chain prompt is used for leading out a logical reasoning process in the large language model, so that the thinking process of the large language model is clarified before the large language model obtains a final answer; and executing the to-be-executed task based on the thinking chain prompt through an executor to obtain an operation result corresponding to the to-be-executed task.