73 results about "Neural coding" patented technology

The invention provides an unmanned cluster brain-like navigation map fusion construction and cooperative positioning method. The method comprises the following steps: acquiring multi-modal perception data, respectively acquiring vision, sound wave and pose information through a binocular camera, a radar and an inertial sensor, simulating a human brain nerve coding mechanism, and generating a pulse sequence and a feature vector in combination with spatio-temporal information; constructing a multi-modal coding unit into a hypergraph node, and based on a dynamic hyperedge connection topological relation, aggregating spatial-temporal characteristics through a heterogeneous hypergraph convolutional network to generate a high-order brain-like semantic map of a single agent; sharing a local brain-like map by multiple agents through distributed communication, detecting geometric and semantic conflicts of an overlapped region, performing space-time alignment based on an anchor point reference, eliminating feature contradictions by utilizing probability distribution matching, and generating a global consistent high-confidence brain-like map; and outputting the optimal position estimation. Through bionic neural coding, heterogeneous hypergraph modeling and multi-agent collaborative optimization, establishment and positioning of a high-order brain-like map in a dynamic unknown environment are realized.

The invention belongs to the field of computer vision, image processing and space engineering, and discloses a lunar soil test piece in-situ three-dimensional reconstruction method based on an improved neural radiation field algorithm, and the method comprises the steps: obtaining a multi-view image set of a lunar soil test piece, and carrying out the self-adaptive adjustment of lunar surface illumination; the pose of the high-resolution camera is calculated through multi-view stereoscopic vision, and a depth map is obtained; the method comprises the following steps: generating a neural coding volume based on a Vision Transform model, and extracting a multi-dimensional feature vector; generating a neural radiation field, and predicting rendering features of the volume points; carrying out micro-volume rendering, and carrying out loss feedback optimization by using color rendering, depth and illumination; performing grid extraction by using a moving cube algorithm to obtain a three-dimensional grid model; for feature extraction difficulty and reconstruction mismatching difficulty caused by lunar soil background texture similarity, the problem of low reconstruction precision under a lunar surface non-uniform illumination condition is effectively solved, and high-quality and high-precision in-situ rapid three-dimensional reconstruction of a lunar soil test piece is realized.

The invention discloses an AI-driven multi-tenant cloud computing resource optimal allocation method and system, and belongs to the field of artificial intelligence, and the method comprises the steps: constructing a tenant resource dynamic portrait, and compressing the tenant resource dynamic portrait into a tensor containing a resource type, a time window and disturbance density through a three-dimensional residual attention network based on a resource request, a response time delay and a load jitter sequence; extracting resource competition conflicts among tenants by using a gating map neural encoder, and outputting a conflict risk map; a hierarchical scheduling space is constructed based on the atlas, a throughput maximization and return optimal strategy is trained, and a periodic switching controller is adopted to realize exploration and convergence balance; carrying out robustness screening on scheduling actions, and eliminating sensitive actions by utilizing a self-adaptive annealing mechanism to generate a stable strategy set; and executing resource allocation and continuously collecting microcosmic utilization data to guide subsequent strategy fine adjustment. The method has the beneficial effects that efficient and stable allocation of resource scheduling is realized, and the resource utilization rate and the system robustness in a multi-tenant cloud environment are improved.

The invention relates to the field of adaptive noise reduction, in particular to an adaptive noise reduction method and system for a multi-mode audio SoC main control chip. The method comprises the following steps: acquiring an original audio input signal according to an SoC main control chip, performing time-frequency domain dual deconstruction and cross-frequency domain noise interference structure analysis, and constructing a full-frequency domain noise interference topology table; carrying out multi-mode interference factor separation on the full-frequency-domain noise interference topology table, carrying out multi-noise environment modeling, and constructing a real-time noise scene model; time sequence noise slope fluctuation modeling is carried out on the real-time noise scene model, dynamic noise reduction response learning optimization is carried out, and a dynamic noise reduction optimization strategy is constructed; and performing neural coding gain and audio boundary line reconstruction according to the original audio input signal to obtain a coding gain key audio signal and a weak audio optimization signal. According to the invention, by flexibly adjusting the noise reduction intensity of the audio signal, the real-time scene noise reduction performance is optimized.

The invention provides a TTS system, a speech synthesis method and device, electronic equipment and a storage medium. The speech synthesis method comprises the following steps: acquiring a phoneme sequence corresponding to a target text; the condition features of the reference speech are extracted, the reference speech is processed into discretized coded representation, and the condition features comprise acoustic features and / or semantic features; inputting the phoneme sequence, the condition feature and the coded representation into a pre-trained speech synthesis model, and obtaining target speech which is output by the speech synthesis model and matched with the target text; wherein the speech synthesis model comprises a neural coding and decoding language model and an audio decoder connected with the neural coding and decoding language model, and the neural coding and decoding language model comprises an autoregression module and does not comprise a non-autoregression module.

Neural coding techniques may be implemented for transmission of redundant audio data. An encoding technique is implemented that uses forward encoding along with an initial state to include multiple audio frames from audio data represented as latent vectors in a network packet transmitted to a recipient. The recipient can then use backward decoding and the initial state to obtain the multiple audio frames from the latent vectors. The multiple audio frames provided in the network packet are redundant audio data that can be used to generate missing audio data.

The invention provides an image compression and image reconstruction method and system based on a single-step diffusion model, and the method comprises the steps: carrying out the feature coding of an input image through a preset neural encoder, and determining the low-dimensional feature representation of the input image; performing compression processing on the low-dimensional feature representation of the input image according to a preset compression ratio, and determining a compression feature representation of the input image; performing a single prediction operation on the compression feature representation of the input image by using a preset single-step diffusion model, and determining an original feature representation of the input image; and decoding the original feature representation of the input image to determine a reconstructed image. According to the method and the device, image compression and image reconstruction in a low-dimensional space and compression feature representation of different compression ratios adopt the same single-step diffusion model to execute a single prediction operation to reconstruct the image, so that the calculation complexity in an image decoding stage is reduced, the image decoding efficiency is improved, and the image structure integrity and the image visual quality are kept.

The invention discloses a brain-computer interface multi-task fine coding and decoding method and system based on electrical stimulation to induce SSSEP, and the method comprises the steps: in a divergent resting somatosensory stimulation test and a somatosensory attention orientation task, applying different frequencies of electrical stimulation and LED stroboflash to the five fingers of a single hand at the same side by using a multi-mode five-finger stimulator, collecting at least six-lead electroencephalogram signals such as C3, Cz, C4, O1, Oz and O2, and carrying out the detection of the electroencephalogram signals; sSVEP features of O1, Oz and O2 leads are analyzed and extracted through a filter bank and are subjected to weighted fusion, and SSVEP classification probability vectors are obtained; the method comprises the following steps: extracting SSSEP classification feature vectors of C3, Cz and C4 leads through a common spatial mode, performing classification to obtain SSSEP probability vectors, finally performing linear weighted fusion on the two vectors, and taking a maximum element to judge a decoding category. According to the method, multi-cortex areas are activated through somatosensory and visual bimodal neural coding, the decoding precision and response speed are improved, the training period is shortened, and rehabilitation medicine is adapted.

A hybrid codec for training a neural coding system, the hybrid codec including a real codec for encoding data pre-processed by a first neural network, and for decoding the encoded data for post-processing by a second neural network; and a proxy codec that is a differentiable representation of the real codec, and is for back-propagating a loss function, representative of a coding bit-rate and a coding distortion between the post-processed data and the original data, for training the weights of at least the first neural network.

The embodiment of the invention provides a video content description method and device, electronic equipment and a storage medium, belongs to the technical field of video processing, and is suitable for the fields of financial science and technology and medical treatment. The method comprises the following steps: acquiring target video data; performing multi-modal data extraction on the target video data to obtain a multi-modal data stream; performing spiking neural coding on the multi-modal data stream to obtain a spatial-temporal feature map; performing hierarchical time sequence transformation on the spatial-temporal feature map to obtain a semantic unit and a Mel-frequency cepstral coefficient; performing cross-modal feature integration on the semantic unit and the Mel-frequency cepstrum coefficient to obtain a cross-modal context representation; and performing text decoding on the cross-modal context representation to obtain target natural language information. According to the embodiment of the invention, the accuracy of video content description can be improved.

The invention discloses a self-adaptive education system based on knowledge gene evolution, and belongs to the technical field of knowledge gene evolution. The system comprises a dynamic knowledge graph engine, a defect response type training sequence generator, a multi-modal memory intensifier and a hypercycle coordination module, wherein the dynamic knowledge graph engine decomposes subject knowledge into heritable units and constructs a dynamic association network of which the weight is updated in real time along with a learning data flow; a defect response type training sequence generator generates a target variable type training sequence according to the cognitive defect node coordinates; the multi-modal memory intensifier generates multi-sensory nerve coding pulses according to training feedback; and the hypercycle coordination module is connected in series with the modules to form a self-adaptive closed loop of knowledge modeling, defect clearing, memory curing and knowledge iteration. The system can generate a cognitive defect thermodynamic diagram, realizes accurate positioning of cognitive vulnerabilities, provides clear targets for teaching intervention, and improves the pertinence and effectiveness of adaptive education.

The invention discloses an extreme image coding and decoding method based on a diffusion model. The method comprises the following steps: S100, a compression and pre-decoding module at a coding end extracts, compresses and transmits image information by using a neural coding network; s200, a compression and pre-decoding module of the decoding end preliminarily decodes the received image information into a content variable aligned with the potential diffusion space by using a neural decoding network; and S300, a denoising module uses the content variable as a control condition and a text graph diffusion model as prior information, and reconstructs an image from random noise through step-by-step denoising. The method can utilize the strong generation capability of the pre-trained text image diffusion model to realize perception-friendly reconstruction consistent with the original image at an extremely low bit rate, has the characteristics of ultrahigh compression ratio and good image reconstruction effect, can carry out image transmission under an extremely low bandwidth condition, and has a wide application prospect. The method can be widely applied to the fields of smart phone satellite communication, short-wave communication, unmanned equipment remote operation and the like.

The present application provides a kind of pulse neural network conversion method and device for underwater sonar autonomous detection, the method comprises the following steps: step S1: modify YOLOv3-tiny model based on deep neural network model (DNN), obtain the YOLOv3-tiny model after modification;Step S2: load underwater sonar image data set, train the YOLOv3-tiny model after modification;Step S3: for underwater target detection, the trained YOLOv3-tiny model is converted into YOLOv3-SNN model based on pulse neural network (SNN), and the YOLOv3-SNN model includes input layer, hidden layer and output layer, wherein input layer uses real number coding, hidden layer uses the two-state pulse neural coding of active state and resting state, and output layer uses membrane voltage decoding;Step S4: using the YOLOv3-SNN model based on pulse neural network (SNN), is directly used for underwater sonar small target detection task.

The invention discloses a method and a system for identifying a connecting line relationship between primitives based on image identification, and relates to the technical field of image processing, the method comprises the following steps: constructing a primitive extraction model, and performing model pre-training in combination with a pre-constructed training data set, the primitive extraction model comprising a primitive identification sub-module and a neural coding sub-module; the wiring diagram to be recognized is input into the primitive extraction model to be analyzed and recognized, a primitive neural coding set is output, and the primitive neural coding set comprises a primitive entity set and an entity coding set; primitive connection relation reasoning is carried out on the primitive neural coding set based on the generative flow network, and a corresponding initial primitive connection network is generated; performing circuit semantic simulation verification on the initial primitive connection network by adopting a pre-constructed differentiable physical simulator to generate a simulation verification result; and iteratively executing the steps S3 to S4 based on a simulation verification result, and generating a precise primitive connection network, thereby realizing efficient, rapid and precise identification of the electrical wiring diagram.

Given prompts including code changes, association intent, and extended context made to a source code program, a code review is automatically generated by a large language model. From the perspective of the code reviewer, the intent represents the problem of code changes and is predicted from a neural classifier given code changes in a code difference format. The neural classifier is a neural encoder transformer model that is pre-trained on various code review datasets and fine-tuned on code disparity blocks of code changes tagged with intent.

The application relates to a large model-based network attack chain timing inference prediction method and system, and belongs to the technical field of network security. The method comprises the following steps: acquiring a plurality of behavior events in a network attack chain, and performing neural coding on the behavior events to map the behavior events to discrete neuron input pulses. Based on different neuron input pulses, synapse connections corresponding to each neuron input pulse are constructed to generate a neuron attack chain graph. The neuron input pulses are integrated by using a dynamic sliding time window, and a large model is called to simulate the charging and discharging behavior of LIF neurons to determine the current state of the behavior events. According to the historical behavior information and the device security state of the behavior events, the current state of the behavior events is analyzed to predict the behavior trend of the network attack chain. The method significantly improves the expression ability of attack chain timing modeling, the prediction accuracy of attack evolution trend, and the initiative and intelligent level of security response.

A hybrid codec for training a neural coding system, the hybrid codec comprising: a real codec for encoding data pre-processed by a first neural network and for decoding the encoded data for post-processing by a second neural network; and a proxy codec being a differentiable representation of the real codec and configured to back propagate a loss function representing the coding bitrate and coding distortion between the post-processed data and the original data, for training weights of at least the first neural network.

The invention is suitable for the technical field of motion intention recognition of electroencephalogram signals, and provides a motion intention recognition method and device based on the electroencephalogram signals and a medium. S1, multichannel electroencephalogram original signals in the user motion imagination process are collected through a dry electrode array; the method comprises the steps of S1, preprocessing signals, S2, extracting time-domain, frequency-domain, time-frequency-domain, space-domain and nonlinear features from the preprocessed signals, and fusing to form a multi-dimensional feature set, S4, performing feature selection and standardization on the multi-dimensional feature set to obtain a standardized feature vector, and S5, inputting the standardized feature vector into a personalized recognition model, outputting a motion intention category, and recognizing the motion intention. S6, an identification result is transmitted to execution equipment, and S7, on the basis of identification performance monitoring, online incremental updating of the model is triggered, a neural coding rule of a motion intention is comprehensively represented through multi-dimensional feature fusion, and the problem that traditional single-domain feature extraction is inaccurate is solved.

A method for encoding a video signal includes receiving encoding instructions for encoding the video signal and encoding the video signal using a training loop for optimizing a metric of the encoded video signal, the training loop including a loss function and one or more of a neural encoder and a neural decoder, and the training loop is adapted based on the encoding instructions.

The invention discloses a body feeling myoelectricity artificial hand system fusing myoelectricity enhanced decoding and bionic electrical stimulation feedback. The system comprises a myoelectricity bracelet, a main controller, an electrical stimulator and an artificial hand. According to the system, an electromyographic bracelet is used for collecting residual limb side electromyographic signals, a main controller removes electrical stimulation artifacts in the signals in real time through a comb filtering and self-adaptive filtering series algorithm, gesture intentions are recognized online through a self-updating multi-class LDA algorithm, and the prosthetic hand is driven to execute corresponding actions. According to the system, a closed-loop feedback mechanism based on a muscle shuttle sensing model is constructed, the muscle shuttle sensing model simulates neural coding characteristics of a human body muscle shuttle, the real-time stroke of a micro servo electric cylinder of the prosthetic hand and stress information of a torque sensor are mapped into electrical stimulation frequency and voltage parameters respectively, and an electrical stimulator is controlled to apply bionic stimulation to the stump. According to the invention, closed-loop control of myoelectricity intention recognition, artificial limb action execution, state detection and electrical stimulation feedback is realized, the problems of lack of sensory feedback, serious electrical stimulation signal interference and the like of the existing myoelectricity artificial limb are effectively solved, a patient is helped to reconstruct ontology perception, and the accuracy and naturalness of artificial limb control are improved.

The invention relates to the technical field of video compression, in particular to a video compression system based on knowledge distillation hardware acceleration, which comprises a data acquisition module, a semantic analysis module, a feature processing module, a compression reconstruction module, a quality evaluation module and a final packaging module. Pixel-level semantic segmentation and region-of-interest recognition are realized through a semantic analysis module, multi-scale spatial-temporal feature extraction and knowledge distillation fusion are performed through a feature processing module, and efficient compression is realized through combination of a cognitive attention mechanism and differentiable neural coding and decoding of a compression and reconstruction module. And performing multi-dimensional quality evaluation and rate distortion optimization through a quality evaluation module, and finally realizing intelligent storage of the code stream by utilizing a final packaging module.

This invention provides a gradient compression method and gradient compressor based on adaptive neural coding, applied to a client-side application, and relating to the fields of distributed machine learning and communication compression technology. The invention obtains the original feature gradient, calculates the importance score of the gradient element to adaptively generate sampling probabilities, and generates a sampling mask through a differentiable sampling mechanism to weight-correct the original feature gradient, obtaining a weighted corrected gradient. Then, it inputs the gradient into a lightweight neural encoder based on a multilayer perceptron architecture to encode a low-dimensional latent representation. A learnable quantization codebook is used to perform soft-allocation quantization on the low-dimensional latent representation, and the output discretized representation is transmitted to the server for reconstruction into a compressed gradient by the server-side decoder. This application can effectively retain key gradient information at a high compression ratio, significantly reduce communication overhead, and maintain model convergence speed and accuracy, making it suitable for resource-constrained scenarios such as federated learning and edge computing.

The application belongs to the technical field of three-dimensional reconstruction and machine learning, and discloses a generalizable neural radiance field reconstruction method based on patch extraction, which is based on a multi-view stereo vision and a volume rendering algorithm of a neural radiance field to implicitly learn a static three-dimensional scene, realize new view synthesis of a complex scene at any angle, and perform three-dimensional reconstruction of the scene. Firstly, two-dimensional image features of a source view and a target view are extracted, and a three-dimensional cost volume is constructed by using a plane scanning algorithm to distort the coordinate system of reference view features through homography transformation. Secondly, the cost volume constructed for the current scene is input into a three-dimensional convolutional neural network to obtain a neural encoding volume. The neural encoding volume is input into a multi-layer perception to regress volume density and color, so as to construct a neural radiance field. Then, the current neural radiance field is used for rendering to obtain a target view, patches of the target view and the reference view are extracted respectively, and reference view patch features and target view patch features are extracted by using a pre-trained VGG-16 low-level network to compare the features. Finally, the average absolute error is calculated to quantify the content feature difference between the target view patch and the reference view patch, and the difference is used as a regular term of an overall loss function to improve the rendering quality of the model. The application can be used for generalizable multi-view three-dimensional reconstruction, can enhance local features through an efficient training mechanism, and can improve the rendering capability for image details and object boundaries.

To provide a neural coding and decoding method, an image coding apparatus, an image decoding apparatus, and a non-transitory machine-readable medium, for more efficient image and video encoding techniques.SOLUTION: A method of image coding comprises the steps of: training a neural network to output pixel data for an image; and outputting the trained neural network as an encoded representation of the image. A method of image decoding comprises the steps of: receiving a neural network trained to output pixel data for an image; and prompting the trained neural network to output the pixel data for the image.SELECTED DRAWING: Figure 4

A hybrid codec for training a neural coding system, the hybrid codec comprising: a real codec for encoding data pre-processed by a first neural network, and for decoding the encoded data for post-processing by a second neural network; and a proxy codec that is a differentiable representation of the real codec, and is for back-propagating a loss function, representative of a coding bit-rate and a coding distortion between the post-processed data and the original data, for training the weights of at least the first neural network.

The invention relates to the technical field of computer-aided diagnosis, in particular to a double-view autism detection method and system based on asynchronous brain function prior. The method comprises the following steps: firstly, acquiring an original fNIRS blood oxygen data sequence and double-view video data of a subject during an experimental task; constructing a global neural encoder by adopting a VGG-like depth one-dimensional convolution architecture, and extracting spatial-temporal features of brain functions to generate a global neural feature vector; constructing a cross-modal channel attention generation network, and generating a channel attention weight vector; injecting the channel attention weight vector into a self-attention module for video feature extraction, and performing channel-level dynamic calibration; and calculating a differential representation vector by adopting a double-flow feature alignment module of time delay perception, and outputting a prediction result of the autism spectrum disorder. According to the method, on the premise that strict time synchronization is not needed, random action noise irrelevant to pathology in the video can be dynamically inhibited, and behavior defect characteristics relevant to neural abnormality can be amplified.

The application discloses a body line correction and nerve remodeling method and system based on electroencephalogram closed-loop feedback, which comprises the following steps: collecting three-dimensional space posture data of a human body in a static or dynamic state, and calculating a force line offset between a current body line and an ideal force line model, so as to adjust the current force line to a target force line state corresponding to the ideal force line model; collecting a brain cortex electric activity signal of a user, obtaining electroencephalogram data, performing feature extraction, obtaining a quantitative index representing a neural coding intensity of the user to the current force line state, mapping the quantitative index into a multi-modal feedback signal, and adjusting a mechanical guiding strategy of a flexible actuator array until the user's brain establishes a stable neural representation of the target force line. According to the technical scheme, multi-modal feedback is used to enhance the somatosensory cognition of the user, the actuator strategy is dynamically adjusted, the brain forms a stable neural representation of the correct force line, and the user can still independently maintain a good body state after leaving the device.

A processing circuit implements: an encoder configured to: supply k symbols of original data to a neural product encoder including M neural encoder stages, a j-th neural encoder stage including a j-th neural network configured by j-th parameters to implement an (nj, kj) error correction code (ECC), where nj is a factor of n and kj is a factor of k; and output n symbols representing the k symbols of original data encoded by an error correcting code; or a decoder configured to supply n symbols of a received message to a neural product decoder including neural decoder stages grouped into a I pipeline stages, an i-th pipeline stage of the neural product decoder including M neural decoder stages, a j-th neural decoder stage comprising a j-th neural network configured by j-th parameters to implement an (nj, kj) ECC; and output k symbols decoded from the n symbols of the received message.

The invention provides a self-adaptive topology decoding test method, device and equipment and a medium, and the method comprises the steps: obtaining neural data collected by a brain under various stimuli, and taking a group of neural data as a sample to construct a data set; respectively decoding the neural coding data of different brain regions in the neural data, and obtaining the decoding precision of the neural coding data of different brain regions according to the decoded classification result; grouping the neural coding data of the different brain regions based on the decoding precision of the neural coding data of the different brain regions; training and testing a self-adaptive topology decoding model by adopting the grouped data to obtain the decoding precision of each group of data; and based on the decoding precision corresponding to each group of data, analyzing the influence of the data in different brain regions on the decoding precision. The technical problem that in the prior art, it is difficult to establish the corresponding relation between the decoding precision of the brain level and the decoding model architecture can be solved and found.

A hybrid codec for training a neural coding system, the hybrid codec comprising: a real codec for encoding data pre-processed by a first neural network and for decoding the encoded data for post-processing by a second neural network; and a proxy codec being a differentiable representation of the real codec and configured to back propagate a loss function representing the coding bitrate and coding distortion between the post-processed data and the original data, for training weights of at least the first neural network.