128 results about "Sequence processing" patented technology

The invention belongs to the field of artificial intelligence and electroencephalogram emotion recognition, and provides a double-branch electroencephalogram emotion recognition method and system based on brain region topology and time-space, and the method comprises the steps: preprocessing a to-be-recognized electroencephalogram signal to obtain a plurality of electroencephalogram fragments, and extracting a difference entropy sequence of each electroencephalogram fragment and a Spearman correlation coefficient matrix between channels; based on the Spearman correlation coefficient matrix, utilizing a bridging dynamic graph attention network module to extract topological features of a brain region; processing the differential entropy sequence by using a multi-scale space-time mixed attention module to obtain multi-scale space-time features; carrying out residual mutual cross attention fusion on the topological features of the brain region and the multi-scale spatial-temporal features to obtain fusion features; and performing classification based on the fusion features, and determining an emotion recognition result corresponding to the electroencephalogram signal. According to the method, the accuracy and robustness of emotion recognition are improved by utilizing the spatial topology characteristics and the multi-topology time dynamic characteristics of the electroencephalogram signals, and the defects of modeling spatial dependence and time dynamic are overcome.

The invention relates to the technical field of signal reconstruction, in particular to a complex electromagnetic environment-oriented millimeter wave signal blind source separation and reconstruction system, which comprises a frequency spectrum trend division module, a path fading construction module, an initial cluster label generation module, a multi-solution path screening module and a fusion reconstruction execution module. According to the method, power spectral density sequence processing is carried out on millimeter wave frequency domain data, a multi-dimensional feature group is formed according to a path loss factor, an angle of arrival and the like, and an initial feature cluster is screened through an Euclidean distance, so that the accuracy of signal source classification is effectively improved; after the frequency domain response and the phase contour are continuously subjected to point comparison, path screening is completed by integrating a mean square error residual value, the path misjudgment probability is reduced, time domain resampling and phase frequency offset standardization are executed in fusion reconstruction, the consistency and fidelity of signal reconstruction are enhanced, and the reconstruction precision is improved. The whole process improves the separation accuracy and reconstruction precision of mixed signals in a complex electromagnetic environment.

One example aspect is directed to a computer-implemented method for performing model decoding with reduced latency. The method includes obtaining a pre-trained sequence processing model comprising a plurality of layers. The method includes modifying the sequence processing model to contain an adapter layer that is configured to receive and process an intermediate representation generated by a particular intermediate layer of the plurality of layers to predict an output token. The method includes training the adapter layer while holding the plurality of layers of the sequence processing model frozen. The method includes deploying the sequence processing model for speculative decoding in which the adapter layer, the particular intermediate layer, and the plurality of layers that precede the particular intermediate layer perform speculative token decoding and the plurality of layers that are subsequent to the particular intermediate layer perform token verification.

The invention belongs to the technical field of artificial intelligence and multi-modal emotion calculation, and discloses a multi-modal emotion recognition method based on a Mama state space model and cross-modal self-distillation. Through the organic combination of the efficient sequence modeling capability of the Mamba state space model and the knowledge sharing mechanism of cross-modal self-distillation, the advantages of the state space model in the aspects of time sequence modeling and calculation efficiency are fully played, and meanwhile, the limitation of a single model architecture is made up through a cross-modal attention mechanism; the technical bottlenecks of an existing multi-modal emotion recognition method in the aspects of long sequence processing efficiency, cross-modal information fusion and knowledge transfer sufficiency are effectively solved, and an efficient and reliable technical solution is provided for further development and practical application of the multi-modal emotion recognition technology.

The invention relates to the technical field of artificial intelligence, and provides a sequence processing method, electronic equipment, a storage medium and a program product.The method comprises the steps that a preset alternative block size set is obtained, and the alternative block size set comprises a plurality of different alternative block sizes; obtaining a sequence length of a target sequence to be processed, and determining a target block size corresponding to the target sequence from the alternative block size set based on the sequence length, the target block size is the alternative block size with the minimum invalid filling data volume generated when the target sequence is subjected to blocking processing in the alternative block size set; and performing attention calculation on the target sequence based on the target block size. According to the method, the optimal block size capable of minimizing invalid calculation is dynamically matched for the sequences with different lengths, and self-adaptive optimization of the calculation load and the sequence length is realized, so that the calculation efficiency and the throughput capacity can be remarkably improved, and meanwhile, the consumption of memory resources is reduced.

The invention discloses a background coherent story picture book generation method based on a diffusion model, and belongs to the field of computer vision and generative artificial intelligence. The method comprises a training stage and a testing stage: in the training stage, bidirectional cross attention fusion and modal soft selection are carried out on text, background and role multi-modal conditions through a feature enhancement fusion module, model optimization is carried out by utilizing joint alignment loss, and efficient parameter fine tuning is carried out on a diffusion model by adopting an efficient parameter fine tuning method; in the test stage, a reference image input by a user and a text sequence are processed into a fusion condition, a large language model is driven to generate an image mark, the image mark is converted into a diffusion condition through a mapper, and each frame of image is generated step by step by combining an autoregression mode with a multi-condition injection diffusion network. According to the method, the problem of inconsistency of cross-frame backgrounds, styles and roles is effectively solved, and high-quality and coherent generation of the long-sequence story picture book is realized.

Computer-implemented methods and systems provide a persistent fixed-size recurrent memory that supports long or unbounded sequence processing with substantially constant compute and bounded memory. A recurrent model maintains a matrix state updated per chunk from key, value, gate, and optional control signals; a gated error between value and a key-weighted state yields a rank-1 proposal. Chunk proposals are reconciled coordinate-wise by an order-invariant convex combiner; solitary proposals pass unchanged. Queries can be answered from proposals without materializing a full updated state. Multiple asynchronous agents emit sparse updates with overwrite strengths that are merged using sparse convex or max selection and optional optimizer preprocessing. A feedback projection writes higher layer activations into lower layer state to consolidate durable skills and reduce forgetting. The approach enables durable retention, parallel inference-time adaptation, scalable linear-attention approximation, and persistent fixed memory use over arbitrarily long sequences.

Provided are systems and methods for fine-tuning sequence processing models (e.g., Large Language Models (LLMs) or Large Multimodal Models (LMMs)) to human preferences. Specifically, provided are systems and methods for application of matching losses, including asymmetric matching losses, at various stages of aligning sequence processing models to reward or preference labels that capture human preferences.

The invention provides a test case generation method and device, equipment, a medium and a program product, and can be applied to the field of artificial intelligence and the field of financial science and technology. The method comprises the steps that the difference between first version core data and second version core data is analyzed, and changed metadata is obtained; on the basis of the change metadata, semantic modeling is conducted on the old-version case data through a test case generation model, and a first test case is generated; wherein the test case generation model is constructed based on a recurrent neural network and an attention mechanism; scheduling and executing the first test case, and collecting multi-modal data in the execution process; processing the test execution data through the exception recognition model to obtain a recognition result; wherein the test execution data comprises multi-modal data, and the anomaly recognition model is constructed based on a sequence processing neural network; and according to the identification result, modifying the first test case based on an exception repair strategy to obtain a second test case.

The application provides an efficient deep learning method and system based on a squeeze-excitation-network and a ConNet network, which comprises the following steps: obtaining a protein global sequence and a protein local sequence as a sample set, and setting model parameters of a DeepNet deep framework; dividing the sample set into a training set and a verification set, setting a model architecture of the DeepNet deep framework, extracting effective feature information, combining negative samples and positive samples in the protein global sequence and the protein local sequence, and sending the samples into the DeepNet deep framework for training and hyperparameter tuning; adaptively coding the protein global sequence and the protein local sequence; designing a local sequence processing branch and a global sequence processing branch, extracting network structure features and key information between long and short sequences by using different scale convolution networks, and calculating final prediction probability according to the key information. The application solves the technical problems that sufficient feature information cannot be extracted and global information is not fully considered and between global information and local information.

The invention discloses a method for realizing lossless switching of agent tasks between basic large models, and relates to the technical field of artificial intelligence. Then analyzing and extracting information from the agent tasks of which the hierarchical categories are determined through an associated metadata template, and carrying out standardization processing to obtain a plurality of standardized agent tasks; processing the standardized agent tasks by a pre-constructed basic large model, recording a processing condition, evaluating and optimizing a basic large model sequence; when an agent task to be processed is processed, the state is processed and monitored according to the optimized basic large model sequence associated with the hierarchical category of the agent task, and the lossless processing flow of the agent task is realized in a mode of lossless switching of the model in the sequence when the state is abnormal.

The application relates to the field of data processing, in particular to a power meter operation quality early warning method based on an intelligent fusion terminal, which comprises the following steps: obtaining total power consumed by a user in each day from a target power meter as daily power consumption, and arranging all the daily power consumptions according to time sequences to obtain an original sequence; processing the original sequence to obtain a first-order difference sequence; presetting an initial value of a cumulative sum, obtaining an improved cumulative sum of the current day by using a composite reference value calculated through the original sequence and the first-order difference sequence, and obtaining an improved decision threshold value through the composite reference value; and performing operation fault early warning of the target power meter in the current day according to the improved cumulative sum and the improved decision threshold value. The application can more accurately detect the operation fault of the power meter, such as metering drift, by comprehensively considering the fluctuation characteristics and scale factors of the power consumption, adaptively adjusting the decision threshold value and the cumulative sum calculation, and reducing false positives and false negatives.

The invention discloses a bit sequence preprocessing method and an OOK symbol generation method and device, and belongs to the technical field of Internet of Things. The method comprises the following steps: acquiring a first bit sequence of which the length is a first number L; under the condition that the first number L is not the integral multiple of the second number M, processing the first bit sequence into a second bit sequence of which the length is a third number L '; wherein the second bit sequence is used for being divided into at least one sequence segment, the length of which is a second number M, for OOK modulation, and M OOK symbols corresponding to each sequence segment are obtained.

A system for controlling an operation of an air-conditioning system including a heat exchanger is provided. The system comprises a processor that executes a neural network trained to simulate an operation of the heat exchanger for a test control input, to produce an output of the simulation based on historical data defining a state of the heat exchanger. The historical data includes a sequence of historical control inputs provided to the heat exchanger and a sequence of historical outputs of the operation of the heat exchanger corresponding to the sequence of historical control inputs. The processor determines a control command to the air-conditioning system based on the predicted test output of the simulation of the operation of the heat exchanger for the test control input and transmits the determined control command to an actuator of the air-conditioning system.

An example method includes generating, using an interface token encoder, a plurality of interface token encodings respectively for a plurality of interface tokens of an input data sequence corresponding to a query. The example method includes generating, using a contextual tokenizer that processes the plurality of interface token encodings, one or more contextual tokens representing the plurality of interface tokens. The example method includes generating, using a machine-learned sequence processing model that processes the one or more contextual tokens, one or more predicted contextual tokens. The example method includes generating, using a contextual detokenizer that processes the one or more predicted contextual tokens, an output sequence including one or more predicted interface tokens that represent the one or more predicted contextual tokens. The example method includes generating, using an interface detokenizer that processes the output sequence, a response to the query.

The application discloses a long context processing robust enhancement method and system based on neuron dynamic adjustment, and belongs to the technical field of text sequence processing, and comprises the following steps: embedding a nerve regulation unit in each Transform layer of a large language model used for a long context processing task; and the nerve regulation unit sequentially performs the following steps: dynamically activating abnormal perception, that is, adaptively calculating an abnormal detection threshold of a current layer according to activation value statistical characteristics of the current layer; activating a hard-gate truncation, that is, generating a binary hard mask according to the abnormal detection threshold, and the binary hard mask realizes deterministic isolation of abnormal components in the activation value while keeping normal components unchanged; and performing hierarchical response compression, that is, performing learnable dimension-level linear compression on the isolated abnormal components, re-integrating the compressed abnormal components and the normal components through gate fusion, and outputting an activation tensor after suppressing the abnormality. The application can prevent rank collapse and gradient oscillation, and enhance the robustness of long context task execution.

The present application relates to a kind of non-woven cloth surface uniformity online detection method, comprising the following steps:1), online image acquisition and pre-processing;2), cloth surface is divided into rectangular detection unit, the gray characteristic value in each detection unit is calculated;3), the unit is numbered respectively along X axis direction and Y axis direction after division;For the gray characteristic value of each X axis, Y axis detection unit, sequence is established along Y axis, X axis direction respectively;4), sequence adopts K-means clustering algorithm to analyze the abnormal cluster of gray characteristic value, and the unit of abnormal value is marked, and the uniformity rate index is calculated according to the total number of abnormal unit;Again, the result after all sequence processing is marked to cloth surface, and the analysis result is visualized;5), according to detection result, trigger multi-stage early warning, realize the online detection of non-woven cloth uniformity.The present application realizes the non-woven cloth from "instantaneous local detection" to "long time overall detection", effectively improves production quality.

This invention discloses a method and system for multivariate time series imputation based on a time-frequency prior enhanced diffusion model, belonging to the field of artificial intelligence and time series data processing technology. The method includes: S1, acquiring the multivariate time series to be imputed and a binary observation mask, separating the observation data; S2, performing preliminary imputation on the multivariate time series to be imputed to form a complete sequence, processing the complete sequence, and generating structured spectral prior features; S3, using the observation data and structured spectral prior features as joint conditions, performing iterative conditional denoising on the missing parts through a denoising diffusion model; and S4, outputting the imputed complete multivariate time series based on the denoising results. This invention, employing the above-mentioned method and system for multivariate time series imputation based on a time-frequency prior enhanced diffusion model, overcomes the technical bottlenecks of existing diffusion models' difficulty in capturing dynamic shifts and frequency domain methods' tendency to smooth high-frequency details, providing a more reliable and complete data foundation for downstream data analysis.

The invention discloses a sequence processing and optimization method and system based on adaptive sparse gating, electronic equipment and a storage medium, and relates to the technical field of artificial intelligence and deep learning. The method comprises the following steps: executing inertial evolution, performing permanent operation with linear complexity by using an inertial processing unit, and updating a hidden state; entropy judgment is carried out, information entropy reflecting prediction uncertainty is calculated, and a gating coefficient is generated; executing on-demand activation, physically bypassing a geometric correction unit at a common time according to a gating coefficient, activating the unit at a critical time, and loading a historical key value pair to perform global self-attention calculation; and finally, manifold fusion is carried out based on the gating coefficient to generate an output state. According to the method, through an adaptive sparse calculation mechanism, the reasoning cost is greatly reduced, and meanwhile, the high precision and anti-forgetting ability of the model in a complex task are effectively guaranteed.

The invention provides an earthquake early warning model, a training method thereof, electronic equipment and a computer storage medium, and relates to the technical field of earthquake early warning. The earthquake early warning model comprises a feature embedding module, a large language model backbone network and a task output module; the feature embedding module is used for performing feature extraction on the three-component strong vibration acceleration waveform data to obtain a feature sequence; the large language model backbone network is used for performing deep semantic feature extraction on the feature sequence; and the task output module is used for outputting earthquake early warning parameters based on the deep semantic features. On the basis of a large language model, through a cross-modal migration technology, the core sequence processing capacity, obtained from the field of'text language ', of the large language model is migrated to the field of'seismic waveform language', and the powerful capacity of large-scale pre-training of the large language model is released. And a low-rank adaptation technology is adopted to carry out fine tuning on the large language model, so that the earthquake early warning model has extremely low training cost and ultrahigh training efficiency.

The invention provides a time sequence prediction method based on deep learning, and relates to the technical field of time sequence prediction, and the method reduces the length of a long sequence through multi-scale decomposition, replaces the traditional full-connection attention through combining with an autocorrelation attention mechanism, prevents the calculation overhead from increasing along with the square of the length of the sequence, remarkably improves the processing efficiency of the long sequence, and improves the accuracy of time sequence prediction. A multi-scale decomposition and bidirectional fusion strategy can simultaneously capture fine-grained fluctuation (the finest scale) and a macroscopic trend (the coarsest scale), the non-stationary data processing capability is superior to that of an existing model, and sequential decomposition (trend-seasonal decoupling) and self-correlation attention (long-range dependence modeling) have a synergistic effect, so that loss of key time sequence information is reduced; the method does not need to depend on manually designed prior hypotheses (such as a fixed period and coarse scale initialization), can adaptively process periodic and non-periodic data, and is wider in application scene.

A computer-implemented method for restoring a sequence for a dataset with frame dropping includes receiving an input sequence. A set of features is extracted from the input sequence. A frequency distribution is determined for the input sequence based on the extracted features. Time domain information for the sequence is restored and in turn, data for the input sequence is augmented based on the restored time domain information. Additionally, noise is removed from the input sequence.

This invention discloses a radar source identification method based on dual-branch fusion and variable-length sequence processing. The method first acquires radar pulse descriptor sequences to construct a standardized dataset. Then, zero-padding is performed on variable-length sequences from the same batch of the dataset, and a binary mask matrix is ​​simultaneously generated to distinguish valid pulses from zero-padding positions. In the feature extraction stage, a temporal branch and an adaptive residual convolutional branch are constructed: the temporal branch utilizes a bidirectional long short-term memory network combined with a length-aware temporal attention mechanism to extract global features; the adaptive residual convolutional branch uses multi-scale convolutional kernels to extract local structural features and employs a mask propagation mechanism to suppress padding interference. Finally, the features from both branches are concatenated and fused, and the prediction result is output through a multi-level fully connected classification network. This invention effectively solves the noise interference problem in variable-length sequence processing and significantly improves the accuracy and robustness of radar source identification in complex environments.

A computer implemented method for generating modified loops for loops with a number of conditionals is provided. A processor set generates a conditional tree based on the number of conditionals in the loops. A value is determined by evaluating each conditional in the number of conditionals The processor set generates a multi-dimensional table based on values from the number of conditionals. The processor set determines an induction variable for the modified loops. The processor set slices the multi-dimensional table to generate a number of slices. The processor set generates a number of sequences by splitting values in each slice. The processor set generates a number of new conditionals based on the determined induction variable and at least values and strides in the number of sequences. The processor set generates the modified loops based on the number of new conditionals.

The invention relates to a biological signal processing technology, in particular to a noninvasive continuous blood pressure estimation system based on a subject adaptive feature modulation mechanism, which is characterized in that an individual prior embedding vector is generated through an individual prior embedding vector generation module; the subject adaptive feature extraction module performs real-time multi-level dynamic modulation on the preprocessed time sequence physiological signal according to an individual prior embedded vector, and outputs a high-dimensional time sequence feature map; the sequence processing module carries out time sequence dependence modeling and context refining on the high-dimensional time sequence characteristic pattern, and outputs time sequence characteristics; the cross-modal attention fusion module splices the individual prior embedded vectors and the time sequence features; the spliced feature vectors are input into a multi-head attention fusion module, dynamic cross-modal feature weighted fusion is achieved, and high-dimensional feature vectors are obtained; and the blood pressure regression output module is used for mapping the high-dimensional feature vector through a full connection layer and carrying out regression to obtain estimated values of systolic pressure and diastolic pressure. The problem that an existing static mapping normal form cannot deal with individual heterogeneity is solved.

The invention discloses a lincRNA small peptide coding potential identification method and a lincRNA small peptide coding potential identification system, and belongs to the technical field of computer technology and bioinformatics. The method comprises the following steps: firstly, processing a lincRNA sequence into a standardized sequence set; identifying and integrating the deduplicated potential coding fragments to obtain a candidate coding region set; after extracting multi-dimensional feature codes of each region, performing fusion through a Mama model and a machine learning model to obtain a fusion prediction confidence coefficient and a label; and finally, judging the coding potential, if not, outputting a conclusion, if yes, deducing an amino acid sequence, calculating comprehensive sequencing confidence by combining multiple indexes, sequencing, and screening small peptide sets with high and low confidence according to a threshold value. According to the method, an efficient sequence state space modeling structure is introduced, dependence on ribosome sequencing or mass spectrum data is not needed, high-precision, low-complexity and extensible prediction of the potential small peptide coding capacity in the lincRNA sequence is achieved, and the technical defect that accuracy and calculation efficiency are difficult to consider at the same time in an existing method is overcome.

The application relates to the technical field of industrial internet, solves the technical problems that a model structure in the prior art is simple, long-range and local dependence cannot be considered, a long sequence is roughly processed, and information loss is large, and in particular relates to an industrial data interaction method, an interaction platform and application of the interaction platform, the platform comprises a preliminary processing module, a preprocessing module, a buffer module, an analysis and prediction module, an interaction module and a security module, the preliminary processing module is used for collecting original data of the industrial data interaction platform, the application can dynamically focus on the importance of different time points in a sequence, no matter how long the dependence span is, while ensuring accurate capture of trend and periodic characteristics, thereby greatly enhancing the expression ability of the model to complex and nonlinear time series relationships, and the application can intelligently compress and retain the most critical information according to the characteristics of an input sequence, so that the model has strong robustness for sequences of different lengths.

Methods and system for predicting complete barcodes from damaged or partially unreadable barcodes at self-checkout (SCO) terminals. A hybrid machine learning architecture is utilized that combines sequence processing for partial barcode data with metadata-based prediction using product attributes such as weight and category information. When a damaged barcode is scanned, available data is extracted and combined with contextual information, and the most likely complete or candidate barcode is predicted. The predicted product details are displayed for customer verification, with the customer given options to approve or reject the prediction. The methods and system continuously improve through feedback logging, with incorrect predictions triggering staff intervention when necessary. This minimizes checkout disruptions, reduces manual intervention, and enhances the overall SCO experience for both customers and retailers.

The invention relates to the technical field of computers, in particular to an event sequence processing method and device, an event query method and device and electronic equipment. Binary coding is carried out on the event field values corresponding to the event fields respectively, and first coding information corresponding to each event field is obtained; obtaining a plurality of fusion fields corresponding to the target event sequence and second coding information corresponding to the fusion fields based on the plurality of event fields of each target event and the first coding information corresponding to the plurality of event fields; splicing the second coding information respectively corresponding to the plurality of fusion fields to obtain target coding information corresponding to the target event sequence; and splicing the target coding information corresponding to the plurality of target event sequences to obtain event coding information. According to the invention, memory occupation of event data can be saved and data transmission efficiency can be improved.

The invention discloses an FFT (Fast Fourier Transform) inverse sequence processing method and device and a radar receiver, which support parallel inverse sequence processing of FFT of multiple paths of data, meet the real-time processing requirement of data, realize efficient completion of inverse sequence conversion of input data and reduce resources. Furthermore, ping-pong cache is used in the system to improve the data access efficiency, and parallel processing between inverted conversion and storage and subsequent butterfly operation is added, so that the processing delay is really reduced, and the performance of the system is optimized.