118 results about "Context vector" patented technology

A multimodal intelligent agent system for dynamic environmental monitoring and user-centered support, consisting of: a multimodal sensor module configured to continuously acquire environmental and behavioral data from multiple input modalities, including at least one visual sensor, at least one acoustic sensor, at least one environmental conditions sensor, and at least one proximity or motion detection sensor, each generating modality-specific data streams representing visual images, audio waveforms, physical environmental parameters, and motion signatures within a monitored environment; a data preprocessing and fusion subsystem that is operationally coupled with the multimodal sensor module and configured to normalize, temporally align, and transform the modality-specific data streams into high-dimensional feature embeddings using a variety of encoders, wherein the visual encoder uses convolutional or vision transformer architectures, the audio encoder uses a spectral-temporal feature extractor, and the sensor encoder transforms raw analog data into context vectors suitable for multimodal alignment; a multimodal processing unit consisting of a transformer-based large language model (LLM) trained on paired multimodal datasets and configured to perform semantic fusion, context abstraction, and inference across the aforementioned aligned multimodal feature embeddings to generate a contextual understanding of environmental and behavioral states; an adaptive agent controller coupled to the multimodal inference processing unit and configured to instantiate, manage, and terminate a variety of task-specific intelligent agents, each agent being a software unit configured to perform a specialized function selected from meeting summarization, behavioral analysis, misplaced object detection, or environmental anomaly identification, with the agents dynamically interacting with the inference engine to retrieve contextually relevant multimodal embeddings for task execution; a personalization and adaptive learning subsystem consisting of a user preference database and a neural memory structure configured to update and refine model parameters based on user-specific interaction history, thereby enabling personalized output generation, prioritization of recommendations, and long-term behavioral adaptation; and An output generation interface is operationally connected to the adaptive agent controller and configured to produce multimodal output in textual, visual, and auditory form. The interface is capable of displaying human-readable summaries, notifications, and visual reconstructions of identified entities or environmental states.

Disclosed in the present invention is a non-autoregressive Transformer-based modeling method for a 4-level pulse amplitude modulation high-speed transmitter. The method involves establishing a deep learning model having an encoder-decoder architecture to predict the behavior of a 4-level pulse amplitude modulation transmitter. An encoder processes unordered non-sequential inputs, including an input signal parameter and link parameters, to generate a context vector and then transmit same to a decoder. The decoder uses both the context vector generated by the encoder and a transmitter output signal sequence to generate a categorical probability distribution for each point in the sequence one by one. The model is trained using a random masking strategy, and inference is performed by means of non-autoregressive decoding and filtering, so that the model can perform parallel prediction on an output sequence, and perform a filtering process to predict an output signal. Compared to traditional simulation methods, the present invention achieves a significant acceleration effect, particularly when processing multi-link systems.

The present application provides a large language model-based query statement generation method, apparatus, and device, and a medium. The large language model comprises a plurality of encoders. The method comprises: acquiring a query text; performing vectorization processing on the query text to obtain a target vector representation corresponding to the query text; inputting the target vector representation into the plurality of encoders, to perform encoding processing on the target vector representation by means of weight matrices of the encoders, so as to obtain context vectors corresponding to the encoders, wherein different encoders have different weight matrices; and generating a target query statement on the basis of the context vectors corresponding to the encoders. A model is allowed to focus on different information features in different attention heads, independently capture different aspects of a query text, and process information at multiple abstraction levels, thereby better capturing complex and abstract semantic relationships, effectively distinguishing the importance of information when processing complex and fuzzy questions, improving the accuracy of query statement generation.

The invention provides a digital human AGI dialogue system based on a cloud side-end collaborative architecture, and relates to the technical field of digital humans, the system is characterized in that a sensing module, a processing module, a decision module, a rendering module and an output module are deployed at a side end, and a decision module, a driving module and a rendering module are deployed at a cloud end; the sensing module collects and preprocesses an input signal of a user; the processing module is connected with the sensing module and is used for extracting features of the input signals and generating context vectors; the decision-making module is connected with the processing module, and generates a decision-making result containing an answer text and an emotion label according to the context vector; the driving module is connected with the decision module, generates an audio stream and a phoneme sequence according to the answer text, and calculates skeleton driving parameters and mouth shape driving parameters of the digital human; the rendering module is connected with the driving module to generate a rendered picture; and the interaction module is connected with the driving module and the rendering module, and aligns the rendered picture and the audio stream to obtain an output result. And low time delay and high performance are realized by adopting cloud edge collaboration.

The invention relates to the technical field of energy prediction, in particular to an energy short-term load prediction method and system based on SE-Block improved Transform. The method comprises the steps of performing reversible normalization preprocessing based on acquired multi-element load sequence data; carrying out feature extraction and fusion on the preprocessed data by utilizing improved cross-scale interaction Patching, wherein the feature extraction and fusion comprise multi-scale feature extraction, cross-scale interaction alignment, residual error correction and dynamic fusion; and performing feature screening on the fused features based on a channel attention mechanism, wherein the feature screening comprises feature response based on improved SE-Block and non-linear interaction of context vectors. Aiming at the non-stationarity of the actual load caused by the influence of meteorological conditions and user behaviors, the model accurately depicts the fluctuation details of the load curve by automatically eliminating the noise interference among multiple variables, and the robustness of the model in the multi-element load prediction of the integrated energy system is reflected.

The invention discloses an attention-based context-aware sparse hybrid expert model routing method, which comprises the following steps of: encoding prompt information input by a user to obtain context vector representation of the prompt information, and introducing a multi-head attention mechanism to obtain multi-scale semantic interaction information; further, constructing a gating network based on attention output, and dynamically selecting Top-K expert networks for reasoning; by introducing a self-adaptive neighbor attention weight and a fusion gating mechanism, expert dispatching and weight fusion of a token level are realized; combining with the sub-output of each expert network, and aggregating according to the weight to obtain the final model output; according to the method, the understanding ability of the model for different semantic contexts is enhanced through a multi-expert structure and a dynamic routing mechanism, and the method is adaptive to multiple rounds of token generation processes, so that the accuracy and diversity of generated texts can be improved.

The invention discloses a front-end cache management method, system and device for a lightweight large-model dialogue state and a medium, belongs to the technical field of front-end cache management of a large-model dialogue system, and aims at solving the technical problem of how to overcome the defects that in a traditional scheme, long context cache is low in efficiency, storage redundancy and insufficient in dynamic semantic adaptation capacity, and the large-model dialogue state cannot be managed easily. In order to realize dialogue context volume compression, improve semantic similar request hit rate and reduce cross-end synchronization delay, the adopted technical scheme is as follows: data acquisition and preprocessing: capturing user interaction behaviors in real time through front-end burying points, and performing preprocessing operation on the acquired user behavior data; semantic normalization processing: performing embedded vector conversion and semantic clustering on the text input by the user to generate a unique semantic identifier and a context vector; querying and updating the multi-level cache; and dynamic collaborative updating: dynamically adjusting the cache based on the cache hit rate, the response delay and the user feedback, and optimizing the cache effect in real time.

The invention relates to the technical field of syndrome evolution prediction, in particular to a syndrome evolution prediction method, system and device based on multi-modal data driving and a medium, the method comprises the steps that multi-modal syndrome data is acquired and preprocessed, and the multi-modal syndrome data comprises clinical symptom data, tongue condition feature data and pulse condition feature data; extracting bidirectional time sequence characteristics in the preprocessed multi-modal syndrome data through a bidirectional long-short-term memory network, and generating a hidden state sequence; weighting the hidden state sequence by using an attention mechanism to obtain context vectors of contribution weights of different time steps; and inputting the context vector into a hidden Markov model of which the state transition matrix is constrained and corrected by the traditional Chinese medicine theory, performing syndrome state reasoning and evolution trend prediction, and outputting a prediction result. The objective of the invention is to improve the prediction precision and interpretability of syndrome evolution.

The invention relates to a power distribution network multi-modal fault reasoning method and device and a medium, and the method comprises the steps: collecting original data, extracting multi-modal features based on the original data, and carrying out the alignment and fusion, and obtaining fusion features; based on the fusion features and the original data, introducing power grid physical rule constraints to construct a space-time knowledge graph; obtaining a text description fault phenomenon, and utilizing the pre-training model to execute the following steps for fault reasoning: carrying out word segmentation on the fault phenomenon, searching a target node in a space-time knowledge graph based on each token obtained by the word segmentation, obtaining a neighbor node of the target node, calculating an attention weight based on the target node and the neighbor node, and carrying out fault reasoning; acquiring a graph context vector based on the attention weight; and performing fault reasoning based on the graph context vector. Compared with the prior art, the capability of handling novel and complex faults is improved.

Systems and methods for Role-Based Access Control (RBAC) using Large Language Model (LLM) embeddings are described. In an illustrative, non-limiting embodiment, an Information Handling System (IHS) may include a processor and a memory coupled to the processor. The memory may store program instructions that, upon execution, generate a plurality of distinct unstructured natural language documents from a same portion of structured data of an enterprise, with each document created for a corresponding role in the enterprise. The IHS may concatenate a role context vector that defines an access privilege for a document with a document vector associated with the document to produce a role-integrated document vector. The IHS may also apply pre-attention and post-attention layers to the role context vector to manage access control during document retrieval based on user roles.

The invention provides a fault detection method and system based on multi-source data fusion, and relates to the technical field of data processing, and the method comprises the steps: obtaining the health state data of a to-be-detected transformer bushing; preprocessing the health state data, and performing spatial feature extraction and fusion on the health state data by using a CNN network to obtain a fusion feature map; determining each pixel point in the fusion feature map as a time step, and capturing time sequence features in the fusion feature map by using a bidirectional long short-term memory network BiLSTM to obtain a time sequence feature sequence; obtaining an attention score of a time sequence feature corresponding to each time step in the time sequence feature sequence, performing weighted aggregation on the time sequence feature according to an attention weight obtained through normalization processing to obtain a context vector, and splicing the context vector and the time sequence feature sequence to obtain time sequence spliced data; and inputting the time sequence splicing data into the trained prediction model, and generating fault prediction information related to the transformer bushing to be detected.

The invention provides a medical document multi-language intelligent analysis system, relates to the field of language translation, and improves the accuracy and efficiency of professional term translation. The method comprises the following steps of: firstly, performing language recognition on an original text document by a recognition module through text unitization and context vector generation, and matching a corresponding corpus; then, a translation module carries out lexical element alignment on the source lexical elements through term bank injection and an AI model, the translation process is automatically optimized, and accurate translation of the terminologies is ensured; and finally, the reconstruction module accurately replaces corresponding contents in the original text document with translation output through the mapping file, so as to ensure that the document format and typesetting are consistent. Through the automatic and optimized translation process, the quality and efficiency of professional term translation are remarkably improved, manual intervention is reduced, and the translation requirement of a high professional standard is met.

The invention discloses a multi-round dialogue logic optimization method in an intelligent question answering system based on a knowledge graph, and belongs to the technical field of artificial intelligence and natural language processing, and the method comprises the following steps: S1, constructing and dynamically updating a triple knowledge graph; s2, based on a graph neural network and a graph attention mechanism, performing time sequence and emotion perception modeling on cross-round semantic association in the triple knowledge graph, and outputting a dialogue context vector; s3, inputting a dialogue context vector and a current user input feature into the generative model, performing semantic mapping through an independent intention, entity and behavior embedding encoder, and generating a robot behavior in combination with logic constraint and emotion-driven weight adjustment from a triple knowledge graph; s4, optimizing the generative model based on a task-associated adversarial training mechanism; and S5, carrying out multi-modal emotion fusion and consistency verification. According to the method, dynamic, coherent and emotional intelligent multi-round dialogue logic optimization can be realized.

The invention relates to an NL2SQL semantic parsing method and device, and the method comprises the steps: inputting a natural language query statement into a business data model, and obtaining a context vector; fusing the context vector with the natural language query statement to obtain a joint semantic vector; decoding the joint semantic vector and the structured feature based on a constraint decoding strategy to obtain an initial SQL statement, and verifying the initial SQL statement through a grammar parser to obtain a target SQL statement; the context vector of the natural language query statement can be more comprehensively captured through the business data model, and the optimal joint semantic modeling is realized through the context vector and the natural language query statement, so that the problems of information bias and redundancy caused by fixed weight or simple splicing are effectively solved; the context features, the structured features of the database and the constraint decoding strategy are dynamically fused, so that complex service logic and anaphora and omission phenomena in the long dialogue context can be better understood.

The invention belongs to the field of natural language processing, and relates to an intelligent memory dynamic evolution method and system based on metadata and two channels, and the method comprises the steps: extracting an entity in an instruction input by a user, and obtaining an instruction entity; querying a metadata index database based on the instruction entity to obtain an entity state; when the entity state is a known state or a non-input state, updating the relational vector database and the graph database through a two-channel mechanism; a user input instruction is matched to the new relational vector database and the new graph database for mixed retrieval, and a preference relation and a hard constraint relation are obtained; constructing a context vector based on the user core portrait, the preference relationship and the hard constraint relationship; processing the user input instruction and the context vector to obtain generated content, and outputting feedback; and the interaction response capability and the engineering landing effect of the large language model intelligent agent in a long-period and complex interaction scene are greatly improved.

The invention discloses a retrieval question and answer method and system combined with a time sequence context, a medium and equipment, and belongs to the crossing field of artificial intelligence and intelligent operation and maintenance of an electric power system. Extracting time sequence semantic elements in the event sequence, and retrieving a corresponding event sub-graph in a time sequence directed graph constructed based on an event sequence, a time sequence and a logic link of the power equipment; performing feature aggregation on the sub-graph nodes to obtain context vectors; performing weighted fusion on the query semantic vector and the context vector to obtain a joint vector; calculating the similarity between the knowledge fragment vector set and each knowledge fragment in the knowledge fragment vector set, and screening out a plurality of most relevant knowledge fragments; and inputting the original query and the knowledge fragments into a text generation model, outputting answers, and completing questions and answers. By implementing the method and the device, the technical problem that an intelligent question and answer retrieval method in the prior art lacks pertinence and professional traceability of retrieval results in a complex event sequence scene in the power field can be solved.

The application discloses a kind of intelligent query control method and electronic equipment based on natural language interaction.The method includes the following steps: building context knowledge base, converting into context vector, and storing in vector database;Convert the input natural language question into query vector;Retrieve the k context vectors most similar to the query vector from the vector database, and determine the corresponding context information;Combine the natural language question and the retrieved context information into a prompt word sequence input into a large language model (LLM) to obtain an SQL query statement, and analyze whether the natural language question contains a control intent;Execute the SQL query statement;If the natural language question contains a control intent, analyze the query result intelligently and perform the corresponding device control operation.The application can analyze the control intent in the natural language question, and after the result is queried, the corresponding device can be controlled according to the control intent to work, improving the work efficiency.

The application discloses a text classification method based on document similarity, and belongs to the technical field of document processing. The method updates the word set of a document sentence through iterative optimization of a word segmentation mechanism and generates a first sentence vector, and generates a first similarity vector in combination with a sentence weight and the first sentence vector; a plurality of theme words are extracted from the word set according to an entity weight and a theme feature vector is generated, a knowledge supplement vector is extracted from a knowledge base based on the context vector of each theme word and the entity weight, a theme enhancement vector is generated in combination with the theme feature vector and the knowledge supplement vector, and a second similarity vector is generated based on the theme enhancement vector; after the first similarity vector and the second similarity vector are compared and optimized in terms of difference, a document similarity vector is obtained and a text category is output. Through iterative word segmentation optimization, theme-guided knowledge enhancement and a similarity comparison mechanism, the application can effectively improve the reliability and accuracy of text classification.

The application discloses a wind power prediction method based on multi-scale space-time coupling, and relates to the technical field of wind energy prediction. The method first collects historical wind speed data of a target turbine and its nearest neighbors, and obtains enhanced features through multi-scale feature extraction and preprocessing; a unique embedding vector is generated for each turbine, which is spliced with the enhanced features to form joint features; the joint features are input into an LSTM encoder to obtain an encoding state, and a spatial context vector is generated through a spatial attention mechanism; enhanced space-time features are obtained through a gating fusion unit, and a multi-layer perceptron is input with a learnable time domain vector of a prediction step to directly regress future 1-12h wind power at one time. The application realizes end-to-end, multi-step and turbine-specific prediction, effectively captures multi-scale space-time coupling rules, reduces prediction error, improves robustness and generalization ability, meets real-time demand of power grid dispatching, and helps large-scale and efficient utilization of wind energy.

A method includes receiving training data including a corpus of multilingual unspoken textual utterances, a corpus of multilingual un-transcribed non-synthetic speech utterances, and a corpus of multilingual transcribed non-synthetic speech utterances. For each un-transcribed non-synthetic speech utterance, the method includes generating a target quantized vector token and a target token index, generating contrastive context vectors from corresponding masked audio features, and deriving a contrastive loss term. The method also includes generating an alignment output, generating a first probability distribution over possible speech recognition hypotheses for the alignment output, and determining an alignment output loss term. The method also includes generating a second probability distribution over possible speech recognition hypotheses and determining a non-synthetic speech loss term. The method also includes pre-training an audio encoder based on the contrastive loss term, the alignment output loss term, and the non-synthetic speech loss term.

The application provides a text data processing method and device, and a training method can include: encoding a word in a target text in a sliding window to obtain a context vector of the word in the sliding window, wherein the sliding window slides on the target text at a preset step; determining a final context vector of the word according to the context vectors of the word obtained by the sliding window in each time passing through the word; and extracting an entity and a relationship between entities from the target text according to the final context vectors of each word in the target text to obtain an entity relationship triple. The method can accurately extract long-distance entities and relationships between entities from the target text to obtain an entity relationship triple.

The application belongs to the field of natural language processing, and relates to an intelligent memory dynamic evolution method and system based on metadata and a double-channel, comprising the following steps: extracting entities in a user input instruction to obtain instruction entities; querying a metadata index library based on the instruction entities to obtain entity states; when the entity states are known states or unrecorded states, updating a relational vector database and a graph database through a double-channel mechanism; matching the user input instruction to the new relational vector database and the new graph database respectively for hybrid retrieval to obtain preference relations and hard constraint relations; constructing a context vector based on a user core image, the preference relations and the hard constraint relations; processing the user input instruction and the context vector to obtain generated content and output feedback; and greatly improving the interactive response capability and engineering landing effect of a large language model agent in a long cycle and a complex interactive scene.

The invention relates to the technical field of man-machine interaction and intelligent wearable equipment, and discloses an emotion recognition and self-adaptive voice interaction method and system in a riding scene, and the method comprises the steps: collecting multi-modal data, such as audio, wind speed, riding mechanics and physiological signals; decoupling the physiological signals by using a personalized physical load model, and constructing a physical load and emotion wake-up vector; fusing physical load, emotion awakening and environmental risk, and constructing a multi-dimensional situation state vector; deciding a service mode based on the context vector, and generating an interaction intention state code; combining the wind speed and the state code to predict wind noise and adaptively filter, and enhancing the voice; generating content according to the service mode and the emotion and adjusting the voice style; and when the user is stable, the personalized physical load model is updated online. According to the method, clear, accurate and safe self-adaptive voice interaction and personalized model calibration during riding are realized by utilizing wind noise dynamic prediction, physiological state decoupling and multi-dimensional situation decision technologies.

This application provides a method, system, and storage medium for predicting the expression efficiency of mRNA cap structures. The prediction method includes obtaining the SMILES sequence of the molecule related to the mRNA cap structure to be predicted; generating a DGL molecular graph based on the SMILES sequence; performing a first encoding on the DGL molecular graph to obtain a graph vector; performing a second encoding on the SMILES sequence to obtain a text context vector; using the text context vector as a query to obtain text-guided graph attention features; using the expanded graph vector as a query to obtain graph-guided text attention features; fusing the text-guided graph attention features and the graph-guided text attention features to obtain a fused feature vector; and performing prediction based on the fused feature vector to obtain the predicted expression efficiency value of the molecule related to the mRNA cap structure. This improves the accuracy of predicting the relationship between mRNA cap structure and expression efficiency.

The invention relates to the technical field of natural language processing, and particularly discloses an English grammar error correction method based on natural language processing, which comprises the following steps: firstly, acquiring and coding an original English text, and generating a context vector sequence containing deep grammar semantics; automatically analyzing and extracting text style features representing the original text writing specification field, and converting the text style features into style condition vectors which can be generated in an adjustable and controllable manner; then, in the process of generating correction candidate texts by autoregression decoding, dynamically injecting the style condition vector into a decoding state and performing attention calculation to realize conditional generation of style perception; and finally, through a multi-dimensional evaluation function of which the weight is dynamically adjusted according to the error type of the original text, performing intelligent preferential selection on the generated candidate results, and outputting a comprehensively optimal correction text. According to the method, a grammar error correction system can automatically adapt to language conventions in different fields, and while the grammar correctness is ensured, the correctness and the result reliability of professional text error correction are remarkably improved.

The application relates to a subtraction gate-based mathematical application question solving method, device, equipment and medium. The method comprises the following steps: encoding a natural language mathematical application question text to obtain a context vector sequence, initializing a starting state vector of a decoder based on the context vector sequence to obtain an initial decoding state vector; iteratively and weightedly fusing the context vector sequence, a historical memory bank and a current decoding state vector through a subtraction gate to obtain an updated decoding state vector; performing mathematical symbol classification based on the updated decoding state vector to obtain generated mathematical symbols; traversing the generated mathematical symbols to obtain a mathematical symbol sequence, constructing a mathematical expression based on the mathematical symbol sequence to obtain a mathematical expression string; and performing evaluation calculation based on the mathematical expression string to generate a solution text corresponding to the natural language mathematical application question text. The method can effectively improve the accuracy of mathematical application question solving.

Embodiments of the present application provide a method and device for determining word representation vectors, electronic equipment and computer readable storage medium, and belong to the field of natural language processing. The method comprises: obtaining a set of word units of a text; obtaining a context vector of the text based on the set of word units; and predicting a next word of the text based on the context vector. The method for determining word representation vectors can effectively obtain a corresponding set of word units even for pictographic characters or languages evolved from pictographic characters that are prone to out-of-vocabulary words, thereby improving the accuracy of determining word representation vectors.

The present application provides a rotary kiln sintering temperature probability interval prediction method, system and device, which introduces a parallel multi-head self-attention mechanism. By fusing the multi-head self-attention feature with the context vector of the sintering temperature prediction model, more abundant information is provided for the decoder, thereby improving the understanding ability of the complex dynamic relationship in the input sequence and enhancing the modeling ability of the long-term dependence relationship. In addition, the present application also adopts a Gaussian process regression (GPR) method to predict the probability interval of the sintering temperature. GPR can infer the posterior distribution and establish a probability prediction model according to the prior distribution of the prediction and the existing data set. Compared with the deterministic method, the GPR prediction contains the uncertainty of parameter estimation, provides a probability prediction interval, and better guides the process control decision.

Aspects of the present disclosure involve reducing latency in generating responses using a large language model (LLM). Aspects include applying a sentence boundary detection technique to an input buffer in order to identify a first sentence in input text while the input text is being input. Aspects include using an encoder portion of the LLM to generate a first embedding of the first sentence. Aspects include determining that a final sentence has been entered in the input text. Aspects include using the encoder portion of the LLM to generate a second embedding of the final sentence. Aspects include generating a cumulative context vector using a merging machine learning model based on the first embedding and the second embedding. Aspects include injecting the cumulative context vector into a decoder portion of the LLM. Aspects include receiving a response to the input text from the LLM based on the cumulative context vector.

The invention discloses a scene character recognition method based on single vector query decoding and related equipment. The method comprises the following steps: performing visual feature extraction on an image to obtain visual features; coding the character to be coded to obtain a character vector; performing global information fusion on the visual features and the character vectors to obtain context vectors; inputting the context vector and the visual feature into a single vector decoder for decoding to obtain a target decoded character; and if the characters in the image are not completely decoded, taking the target decoded character as a to-be-coded character, and returning to the step of coding the to-be-coded character by adopting the position sensing hash coding method to obtain the character vector until the characters in the image are completely decoded, thereby obtaining a scene character recognition result. According to the method, the recognition accuracy of the model on shielded and fuzzy samples can be improved, the problems of attention drift and calculation amount increase faced by a traditional autoregression decoding method can be relieved, and the method can be widely applied to the technical field of computer vision.