Performing network congestion control utilizing reinforcement learning

Obstacle trajectory prediction method and device, equipment and storage medium

The invention relates to an obstacle trajectory prediction method and device, equipment and a storage medium, and particularly relates to the technical field of intelligent driving. The method comprises the following steps: determining a target obstacle in a specified range of a target vehicle; determining interaction map sequence information and historical state information of the target obstacle; determining a cross splicing fusion feature sequence of the target obstacle according to the target obstacle interaction map sequence information and the historical state information; and according to the cross splicing fusion feature sequence of the target obstacle, processing through a recurrent neural network model, and determining a target trajectory of the target obstacle. According to the scheme, all the motion information of the target obstacle and the position and speed relation between the target obstacle and other obstacles are considered at the same time, the data are fused and then ...

Pulse neural network architecture based on liquid state machine

The invention relates to a liquid state machine-based spiking neural network architecture, which comprises a storage layer neuron array and a learning layer neuron array which are respectively used for simulating a storage layer and a learning layer, the storage layer neuron array comprises a weight generation and probability transmission module, the weight generation and probability transmission module is configured to be provided with a first linear feedback register and a second linear feedback register, when a presynaptic pulse comes, a random number seed in a specified random number seed memory is read, so that the first linear feedback register generates a fixed random number sequence, and the second linear feedback register generates a random number sequence; generating weights of the post-synaptic neurons and different pre-synaptic neurons; the learning layer neuron array comprises an IF neuron module, a membrane potential storage circuit, a weight storage circuit and a weight learning ...

Method and device for predicting time sequence index

The embodiment of the invention provides a method and device for predicting a time sequence index. According to the method, first, corresponding n first representations are obtained based on respective service features of n service objects in a target time period t; wherein the n business objects form nodes in the tree structure based on the range affiliation relation of the n business objects. Thirdly, performing convolution operation on the n first representations to obtain n second representations; wherein the second representation corresponding to any target business object in the n business objects is obtained by performing convolution operation on the first representation in a target convolution window, and the target convolution window comprises a plurality of business objects covered by a path from the root node to the target business object in the tree structure. Based on this, n object representations of the n business objects can be determined according to the n second representations ...

Airport terminal fresh air energy-saving optimization control method based on passenger space-time distribution prediction

The invention relates to the technical field of intelligent control of an airport terminal environment control system, in particular to an airport terminal fresh air energy-saving optimization control method based on passenger space-time distribution prediction, and the method comprises the following steps: S1, based on chi-square distribution, building an airport terminal passenger harbor abutting probability model, and applying a fluid dynamics principle to obtain a passenger harbor abutting probability model; passenger flow space-time distribution prediction of an airport terminal passenger flow line is realized; s2, establishing an airport terminal indoor air quality prediction model based on a gating circulation unit network and a passenger space-time distribution prediction result; and S3, based on the fuzzy control theory and the indoor air quality prediction result, a fresh air fuzzy control strategy of the air conditioning system of the airport terminal building is formed, and ...

Speech recognition method, system, device and storage medium

The invention discloses a speech recognition method, a speech recognition system, speech recognition equipment and a storage medium, which are used for acquiring original speech and transmitting the original speech to a trained Adam-BP neural network so as to output a speech recognition result. Wherein the training process of the constructed Adam-BP neural network is as follows: S1: carrying out initialization setting on an Adam-BP neural network model; s2, collecting a training sample set, and inputting the training sample set into the Adam-BP neural network, S3, defining an objective function of the Adam-BP neural network; s4, using the objective function to perform back propagation to update the parameter vector, applying the parameter vector to the Adam-BP neural network, circularly entering the step S2, and performing iteration to update the parameter vector until a convergence condition is satisfied; according to the voice recognition method, intelligent voice recognition, including ...

Equipment detection model construction method and device, computer equipment and storage medium

The invention relates to an equipment detection model construction method and device, computer equipment and a storage medium. The method comprises the steps of obtaining device operation data of a target device, obtaining feature data of each dimension according to morphological features of a target object of infrared image data, determining the feature data of each dimension as target feature data in sequence, determining a corresponding weight matrix according to the target feature data, and obtaining a weight matrix corresponding to the target feature data; and fusing the feature data of each dimension with the corresponding weight matrix to generate each sample data, training according to each sample data to obtain each data enhancement model and corresponding enhancement data, determining target enhancement data according to the similarity between the enhancement data corresponding to each data enhancement model and the infrared image data, and outputting the target enhancement data ...

Subsynchronous oscillation identification method based on environmental excitation data

The invention discloses a subsynchronous oscillation identification method based on environmental excitation data, and relates to the technical field of power system stabilization and control, and the method comprises the following steps: receiving environmental excitation data in a power system; inputting the environmental excitation data into a pre-established gated circulation unit neural network GRU model, and extracting subsynchronous oscillation characteristics of the environmental excitation data; identifying subsynchronous oscillation characteristics by using an Attention mechanism, and obtaining frequency and damping ratio parameters of subsynchronous oscillation through full-connection layer dimension transformation; the GRU model can realize preliminary extraction of subsynchronous oscillation modal characteristics in environmental excitation signals, and the Attention mechanism can make the model selectively focus on specific time sequence characteristics, so that the information ...

Method for predicting putting amount of algistat for treating cyanobacterial bloom

The invention provides a method for predicting the putting amount of an algistat for treating cyanobacterial bloom, which comprises the following steps of: 1, collecting an algae-containing water sample, and setting an experimental group and a control group; 2, inputting the algistat putting concentration and the measured initial water quality index as a model; 3, establishing a BP neural network model optimized based on a particle swarm and a genetic algorithm in order to reduce an error between a predicted water quality index and a real water quality index under different algistat putting concentrations; 4, the putting concentration is changed, the trained model is used for predicting the water quality after putting and judging whether the water quality reaches the standard or not, the water quality prediction speed is greatly increased, the prediction precision of the putting amount of the algistat can be remarkably improved, and the method is suitable for popularization and application ...

Data processing method and device, readable medium and electronic equipment

The embodiment of the invention relates to a data processing method and device, a readable medium and electronic equipment. The method comprises the steps of obtaining to-be-processed business data, obtaining a plurality of first target features corresponding to the business data according to a pre-generated target feature extraction rule, inputting the plurality of first target features into a pre-generated event recognition model, and obtaining a target recognition result output by the event recognition model. Wherein the target feature extraction rule is a rule generated according to first sample data, a preset feature extraction rule and a preset feature derivation rule, the preset feature extraction rule is used for obtaining a first sample feature of the first sample data, and the preset feature derivation rule is used for performing derivation calculation on the first sample feature; obtaining candidate derivative features; the event recognition model is a model obtained by training ...

Park energy big data management method and system based on machine learning

The invention discloses a park energy big data management method and system based on machine learning, and the method comprises the steps: employing a Monte Carlo simulation method, and randomly selecting a preset number of arbitrary values in a preset interval for simulation based on a time vector sequence of park energy; training a data set for the simulated sample by using a Bayesian loop network, generating a target network, judging posterior distribution of the weight of the target network, and carrying out network updating iteration; and controlling switching of energy types based on an output layer classification result of the Bayesian cycle network. The method does not need to explicitly specify a probability model or fit probability distribution characteristics, and due to the characteristics of the data model, it is determined that the method has the convenience of not needing sampling steps and manually marking data, and the accuracy is ensured. In addition, a Bayesian formula ...

Processing apparatus and inference system

According to an embodiment, an inference system includes a recurrent neural network circuit, an inference neural network, and a control circuit. The recurrent neural network circuit receives M input signals and outputs N intermediate signals, where M is an integer of 2 or more and N is an integer of 2 or more. The inference neural network circuit receives the N intermediate signals and outputs L output signals, where L is an integer of 2 or more. The control circuit adjusts a plurality of coefficients that are set to the recurrent neural network circuit and adjusts a plurality of coefficients that are set to the inference neural network circuit. The control circuit adjusts the coefficients set to the recurrent neural network circuit according to a total delay time period from timing for applying the M input signals until timing for firing the L output signals.

Flexible parameter sharing for multi-task learning

Systems and methods for flexible parameter sharing for multi-task learning are provided. A training method can include obtaining a test input, selecting a particular task from one or more tasks, and training a multi-task machine-learned model for the particular task by performing a forward pass using the test input and one or more connection probability matrices to generate a sample distribution of test outputs, training the components of the machine-learned model based at least in part on the sample distribution, and performing a backwards pass to train a connection probability matrix of the multi-task machine-learned model using a straight-through Gumbel-softmax approximation.

Method and system for implementing a natural language interface to data stores using deep learning

An embodiment of the present invention is directed to combining natural language processing with constrained grammar defined pattern matching to advantageously allow a system to translate natural language questions into any number of underlying technologies. By using a unique intermediate parse tree representation, the disclosed embodiments are able to instantiate a corresponding data store adapter for each given query, which may be for a relational database or a no-SQL database, for example. The ability to abstract underlying storage technology advantageously allows the management of disparate database systems, which is not possible using existing methods and technology.

COMPUTATIONAL STORAGE FOR AN ENERGY-EFFICIENT DEEP NEURAL NETWORK TRAINING SYSTEM

A training system includes a dynamic random access memory (DRAM) configured to buffer training data; a central processing unit (CPU) coupled to the DRAM and configured to downsample the training data and provide the DRAM with the downsampled training data; a computational storage consisting of a solid-state drive (SSD) and field-programmable gate array (FPGA) and configured to perform dimensionality reduction on the downsampled training data to generate training data batches; and a graphic processing unit (GPU) configured to perform training on the training data batches.

Neural network-based message communication framework with summarization and on-demand audio output generation

Methods, apparatus, and processor-readable storage media for a neural network-based message communication framework with summarization and on-demand audio output generation are provided herein. An example computer-implemented method includes obtaining message communication content; determining intent-related information and domain-related information in the obtained message communication content by processing at least a portion of the obtained message communication content using one or more machine learning-based natural language processing techniques; generating a summarization of the obtained message communication content by processing, using at least one neural network, at least a portion of the obtained message communication content in connection with at least a portion of the determined intent-related information and at least a portion of the domain-related information; converting the generated summarization from a text format to an audio format; and performing at least one automated ...

INSTRUCTIONS AND LOGIC TO PERFORM FLOATING-POINT AND INTEGER OPERATIONS FOR MACHINE LEARNING

The present disclosure provides a data processing system, a method, a computer-readable medium and a graphics processing unit, GPU, to accelerate machine-learning operations, the GPU comprising: a multiprocessor including a single instruction, multiple thread, SIMT, architecture, the multiprocessor to execute a single instruction across multiple threads; and a first compute unit included within the multiprocessor, the single instruction to cause the first compute unit to perform at least a two-dimensional matrix multiply and accumulate operation, wherein to perform the two-dimensional matrix multiply and accumulate operation includes to compute an intermediate product of 16-bit operands and to compute a 32-bit sum based on the intermediate product; wherein to compute a 32-bit sum based on the intermediate product, the first compute unit is to: perform a floating-point multiply of two or more 16-bit operands to generate the intermediate product, compute an intermediate sum based on the intermediate ...

Exposure defects classification of images using a neural network

A method 300 of training a neural network to classify image exposure defects comprises: receiving a first set 302 of labelled images; training a neural network 304 using the labelled images; receiving a second set 306 of unlabelled images; applying the second set of images to the neural network to produce inference scores 308 based on the exposure quality; classifying each image of the second set as well-exposed or poorly exposed based on the inference scores; and retraining the neural network model based on the classifying and the second set of images. The first set of images are each labelled as either well-exposed or defective. The classifying is carried out based on a distribution ratio of well-exposed and defective images in the first set in addition to the inference scores. The defects may be underexposure or overexposure. The retraining may comprise determination, based on the distribution ratio and inference scores, that an image of the second set is misclassified, and retraining ...

Detection of phishing campaigns

Information pushing method based on electronic commerce big data

The invention discloses an information pushing method based on electronic commerce big data. The method comprises the following steps: acquiring data information of a target user; constructing a target user interest model, and calculating to obtain a first interest set and a second interest set; based on a BP neural network, taking the first interest set and the second interest set as input nodes to obtain a final push result; and pushing the push result to the target user. According to the method, the target user interest model is constructed on the basis of various data of the e-commerce platform, so that real target user interests and preferences can be mined, accurate pushing can be carried out on the target user, the requirements of the target user can be met according to the user requirements, the search difficulty of the user for searching favorite commodities is reduced, and the user experience is improved. The shopping experience of the user using the e-commerce platform and the ...

Irregular text tagging method

The invention relates to the technical field of text processing, in particular to an irregular text tagging method, which comprises the following steps of: when the irregular text is tagged, firstly judging the irregular text, judging whether the irregular text is a long text or a short text, and when the irregular text is the long text, carrying out tagging on the irregular text; when the irregular text is a short text, labeling the long text by adopting a deep learning model, and when the irregular text is the short text, labeling the short text by adopting a sample-based classification model; in this way, the classification accuracy of irregular texts can be effectively improved.

Integrated learning electricity price prediction method and device based on distributed data fusion

The invention relates to an integrated learning electricity price prediction method based on distributed data fusion, and the method comprises the steps: collecting historical electricity price data of each city distributed in a provincial company region; performing data preprocessing on the historical electricity price data to obtain processed data; training the processed historical electricity price data through a machine learning model, and then performing electricity price prediction to obtain an initial prediction result of the electricity price; verifying the model after machine learning by using the processed historical electricity price data to obtain the total error of each city, and performing weighted fusion on the preliminary prediction result to obtain the future predicted electricity price of each city; and each city uploads the respective future predicted electricity price to the center for decision level fusion to obtain the future predicted total electricity price of the ...

Classification component information processing application system

The invention relates to a classified component information processing application system, and the system comprises a customization detection mechanism which is used for obtaining a red and green component value, a black and white component value and a yellow and blue component value corresponding to each pixel point of an input image in an LAB space, and further identifying the targeted sharpening level of the input image; and the target judgment equipment is used for intelligently identifying the number of human body targets in the input image based on the targeted sharpening level of the input image, the contrast ratio of the input image and the field depth numerical values corresponding to the pixel points of the input image by adopting a recurrent neural network. According to the invention, the number of human body targets in the image can be intelligently identified based on the targeted sharpening level and contrast of the image and each depth-of-field value corresponding to each ...

Method and device for detecting position relation between sight line and observation point

The invention discloses a method for detecting a position relationship between a sight line and an observation point. The method comprises the following steps: acquiring an experiment video frame containing the observation point and the face of an observer; inputting the experiment video frame into the trained target detection model, and outputting a target detection result containing a human eye detection frame and an observation point detection frame; performing iris segmentation on the eye region based on a human eye detection frame in the target detection result to obtain an iris segmentation result; and based on the target detection result and the iris segmentation result, judging whether the observer horizontally reads the reading of the observation point or not. According to the scheme, whether an observer horizontally reads the reading or not in the experiment process can be more accurately judged in real time.

Property equipment management method and system based on big data visualization

The invention discloses a property equipment management method and system based on big data visualization, and relates to the technical field of visual management, and the method comprises the steps: collecting the related data of equipment in a property system; performing deep analysis and prediction on the related data through a convolutional neural network and a recurrent neural network, generating a first risk coefficient of the equipment, and performing primary judgment; triggering a corresponding equipment management strategy according to the primarily judged risk level; and obtaining equipment state data again, performing second risk prediction through the deep learning model in combination with a historical risk prediction value of the equipment to obtain a second risk coefficient, calculating a standard deviation, and performing final judgment according to the second risk coefficient and the standard deviation. According to the method, the equipment state is displayed and the equipment ...

Semantic text similarity calculation method for multi-angle enhanced network

The invention relates to a semantic text similarity calculation method for a multi-angle enhanced network, which comprises the following steps of: performing word segmentation on a Chinese text, obtaining text representation by using character embedding and word embedding, and training character embedding and word embedding in a corpus; fusing character embedding and word embedding to obtain a fusion vector, and encoding the fusion vector by using Bi-SRU; further capturing the similarity and difference of the two texts by using soft alignment attention; global similarity modeling is carried out on the text, and richer global similarity information is captured; representing the result vector as a fixed vector through a pooling layer, and inputting the fixed vector into an MLP multi-classifier for classification; and predicting the similarity degree of the two texts through softmax of the MLP multi-classifier, and outputting a prediction probability. The problems that rich Chinese semantics ...

Patent text entity extraction method and device, equipment and medium

The invention discloses a patent text entity extraction method and device, equipment and a medium, and the method comprises the steps: generating at least one triple according to a claim subject, a claim reference relationship and a claim serial number in a claim text, and generating a reference relationship topological graph corresponding to the claim subject according to the triple; inputting the claim text corresponding to each reference relation topological graph into a pre-training model bert to generate a first entity pair; dividing the claim text corresponding to each reference relation topological graph into at least one technical scheme block according to the reference relation, and respectively inputting the technical scheme blocks into a pre-training model bert to generate corresponding technical scheme entity pairs; splicing the technical scheme entity pairs to generate a second entity pair; and cutting and fusing the first entity pair and the second entity pair to generate ...

Social event abstract generation method and device, terminal equipment and medium

The invention is suitable for the technical field of artificial intelligence, and provides a social event abstract generation method and device, terminal equipment and a medium, and the method comprises the steps: obtaining graphic and text information of a target event; determining an event type of the target event according to the graphic and text information; determining an abstract key point label of the target event according to the event type and a preset event body schema; determining event key point information and knowledge graph information of the target event according to the abstract key point tag; and inputting the graphic and text information, the event key point information and the knowledge graph information into a trained abstract generation model, and inputting the abstract key point tag into the abstract generation model, so that the abstract generation model generates abstract content corresponding to the abstract key point tag, and then an abstract corresponding to ...

Full-color-gamut structural color prediction and reproduction method based on multi-level micro-nano structure

The invention discloses a full-color-gamut structural color prediction and reproduction method based on a multi-level micro-nano structure. The method comprises the following steps: firstly, establishing a multi-dimensional mapping relation between a structural color primary color proportion and a predicted color spectral reflectivity, and then constructing a structural color color prediction forward model and a structural color proportion prediction reverse model to realize full-color-gamut color prediction and reproduction of any color and any tone, which is still the first time in the field of structural color research. Wide application prospects are realized.

Method and an apparatus for determining an allocation decision for at least one elevator

A method for determining an allocation decision for at least one elevator includes using an existing calls in an elevator system as a first input in a machine learning module, processing the first input with the machine learning module to provide a first output comprising a first allocation decision, using the first output as a second input in an iterative module, processing the second input with the iterative module to provide a second ouput comprising a second allocation decision, and providing the second allocation decision to an elevator control module and to an allocation decision storage for further machine learning module training.

Auto-generation of subtitles for sign language videos

Embodiments are disclosed for a subtitle generator for sign language content in digital videos. In some embodiments, a method of subtitle generation includes receiving an input video comprising a representation of one or more sign language gestures, extracting landmark coordinates associated with a signer represented in the input video, determining derivative information from the landmark coordinates, and analyzing the landmark coordinates and the derivative information by at least one gesture detection model to identify a first sign language gesture.

ARTIFICALLY INTELLIGENT MEDICAL PROCEDURE ASSESSMENT AND INTERVENTION SYSTEM

A system and methods for artificially intelligent medical procedure assessment and intervention are provided. A system may acquire image data from a camera targeting a location where healthcare is administered. The system may receive sensor data from a sensor attached to a care provider during performance of a procedure. The system may generate gesture features based on the sensor data. The system may generate image features based on the image data. The system may determine, based on the image features and a machine learning model, a step identifier for a step in a multi-step surgical procedure. The system may access text descriptive of an instruction to perform the step. The system may display the text on a display accessible to the care provider. The system may determine, based on the gesture information and a second machine learning model, a performance metric that measures performance of the care provider.

Protein database search using learned representations

A method for efficient search of protein sequence databases for proteins that have sequence, structural, and/or functional homology with respect to information derived from a search query. The method involves transforming the protein sequences into vector representations and searching in a vector space. Given a database of protein sequences and a learned embedding model, the embedding model is applied to each amino acid sequence to transform it into a sequence of vector representations. A query sequence is also transformed into a sequence of vector representations, preferably using the same learned embedding model. Once the query has been embedded in this manner, proteins are retrieved from the database based on distance between the query embedding and the protein embeddings contained within the database. Rapid and accurate search of the vector space is carried out using exact search using metric data structures, or approximate search using locality sensitive hashing.

Managing off-grid energy generation and storage

Techniques for managing an off-grid power system include executing update requests for an off-grid power system that is communicably coupled to an energy management system by determining an amount of stored energy in energy storage devices in response to at least one update request, determining an amount of electrical power generatable by renewable energy power systems in response to another update request, determining a predicted reliability of at least a portion of the energy storage devices and the renewable energy power systems in response to another update request, and determining an amount of electrical power for a remote facility that is electrically coupled to the off-grid power system in response to another update request. The techniques further include determining a control command for the off-grid power system based on the responses to the update requests; and providing the control command to the off-grid power system to adjust an operation of at least one of the energy storage ...

System and method for language translation

Provided are computer implemented systems and methods for providing improved language translation, including translation of input text to an output translated text element. This may include providing for rule-based language translation. This may further include a plurality of machine translation models adapted to different translation domains. A machine translation selector is described that selects an in-domain machine translator from the plurality of machine translation models based on an input text element. The selected in-domain machine translator may generate a first translated text element from the input text element. A post-editor module may generate a second translated text element by predicting a post-edit to the first translated text element. The first translated text element and the second translated text element are evaluated by a quality evaluation model that determines a first metric associated with the first translated text element and a second metric associated with the ...

Real-time alerts and transmission of selected signal samples under a dynamic capacity limitation

Real-time alerts and transmission of selected signal samples is disclosed. One example is a system including a base facility linked to a production station with an alerting system to perform anomaly analysis utilizing an anomaly model. A receiver at the base facility receives, from the production station, a selection of signal samples based on the anomaly analysis, where the received selection is optimized at the production station to be substantially relevant to an update of a statistical model while adhering to a dynamic capacity limitation of the production station. The statistical model is maintained at the base facility and incorporates features related to the production station. A management system at the base facility updates the statistical model based on the received selection, optionally derives an updated anomaly model based on the statistical model, and optionally transmits the updated anomaly model to the production station.

Information processing device and information processing method

There is provided an information processing device which efficiently executes machine learning. The information processing device according to one embodiment includes: an obtaining unit which obtains a source code including a code which defines Forward processing of each layer constituting a neural network; a storage unit which stores an association relationship between each Forward processing and Backward processing associated with each Forward processing; and an executing unit which successively executes each code included in the source code, and which calculates an output value of the Forward processing defined by the code based on an input value at a time of execution of each code, and generates a reference structure for Backward processing in a layer associated with the code based on the association relationship stored in the storage unit.

Fog computing systems and methods

Fog computing systems are provided comprising edge-nodes and middle-nodes between edge-nodes and cloud-node. These nodes form a hierarchical structure with each cloud, middle node having children nodes, and each middle, edge node having a parent-node. Each edge-node receives data from sensors, assigns reception-timestamp to each data indicating when data has been received to produce series of timestamp-ordered data, trains local model through machine-learning based on said series of timestamp-ordered data, and sends said series to parent-node of the edge-node. Each middle-node collects series of timestamp-ordered data from children nodes of the middle-node, trains supra-local model through machine-learning based on said collected series of timestamp-ordered data, and sends said collected series to parent-node of the middle-node. Parent-children structures, edge-nodes and middle-nodes for such fog computing systems are also provided. Methods and computer programs of operation of an edge-node ...

Setting latency constraints for acoustic models

Methods, systems, and apparatus, including computer programs encoded on computer storage media for acoustic modeling of audio data. One method includes receiving audio data representing a portion of an utterance, providing the audio data to a trained recurrent neural network that has been trained to indicate the occurrence of a phone at any of multiple time frames within a maximum delay of receiving audio data corresponding to the phone, receiving, within the predetermined maximum delay of providing the audio data to the trained recurrent neural network, output of the trained neural network indicating a phone corresponding to the provided audio data using output of the trained neural network to determine a transcription for the utterance, and providing the transcription for the utterance.

LARGE-SCALE MULTILINGUAL SPEECH RECOGNITION WITH A STREAMING END-TO-END MODEL

Communication method and device

The invention provides a communication method and device, which can reduce the transmission overhead of training data and improve the performance of an artificial intelligence model. The method comprises the following steps: acquiring first data and first information from a first node, wherein the first information is used for indicating a data enhancement mode of the first data; determining a first training data set of the model according to the first data and the first information; ...

Cross-domain pedestrian re-identification method

The invention discloses a cross-domain pedestrian re-identification method, which comprises the steps of S1, learning an image style of a target domain through a style conversion model, so that image data in a source domain can simulate the target domain after conversion; and S2, a pedestrian re-identification task is completed, and a pedestrian re-identification network model is improved based on a ResNet50 network, so that the extracted features are more robust. The inter-domain difference is reduced by using the cyclic generative adversarial network, and the recognition effect of the pedestrian re-recognition network is improved.

Heating furnace temperature optimization method based on hybrid modeling and improved sand dune cat swarm algorithm

The invention provides a heating furnace temperature optimization method based on hybrid modeling and an improved sand dune cat swarm algorithm. The method comprises the steps that 1, billet parameters, heating furnace structure parameters, production working conditions and operation parameters are collected; 2, establishing a billet temperature distribution hybrid forecasting model, and predicting billet temperature distribution at the current moment according to billet temperature distribution predicted at the previous moment, heating furnace structure parameters, and furnace temperature and production working conditions and operation parameters at the current moment; 3, a heating furnace temperature optimization objective function is constructed, and constraint conditions are constructed based on billet temperature distribution predicted by the hybrid forecasting model and charging and discharging temperatures; and 4, according to the rolling rhythm, an improved SCSO algorithm is adopted ...

Game super-resolution method based on recurrent neural network

The invention discloses a game super-resolution method based on a recurrent neural network, and belongs to the technical field of super-resolution. According to the method, an input structure is constructed by using an estimation result of a previous frame of low-resolution image, an estimation result of a previous frame of high-resolution image and hidden layer information transmitted by a previous frame of recurrent neural network, the high-resolution image of the current frame is estimated through a neural network model, and the hidden layer information of the current frame of recurrent neural network is updated; and then the generated information is transmitted to the next frame to start the next cycle. Through continuous circulation of the process, based on the game super-resolution method provided by the invention, a frame of low-resolution game picture can be input, a frame of high-resolution game picture can be output, the information of the game picture sequence in the time domain ...

Multi-source information propagation prediction method and system based on heterogeneous graph neural network

The invention provides a multi-source information propagation prediction method and system based on a heterogeneous graph neural network, and the method comprises the steps: constructing a heterogeneous graph based on social network information; generating an initial feature for each class node and edge in the heterogeneous graph through a set initial feature generation strategy based on the heterogeneous graph; inputting the heterogeneous graph and the initial features into a pre-constructed multi-layer space-time embedded network model for optimization training to obtain a trained multi-layer space-time embedded network model; and predicting multi-source information propagation in the social network by using the trained multi-layer space-time embedded network model. According to the method provided by the invention, the defects that a traditional model is lack of flexibility in a time period division scheme, redundant in attention relation, incapable of giving consideration to spatial-temporal ...

Metering master station fee control module encryption method fusing federated learning and swarm intelligence

The invention discloses a federated learning and swarm intelligence fused metering master station cost control module encryption method. The method comprises the steps that a metering master station server constructs a cost control data training model; encrypting the data of the cost control function module by using a differential privacy protection method, and evaluating the minimum privacy loss; and the model parameters uploaded to the intelligent electric meter terminal are aggregated, and the aggregated model parameters are downloaded to the local marketing terminal to be decrypted and the local model is updated. According to the metering master station cost control module encryption method fusing federated learning and swarm intelligence, the terminal parameters of the intelligent electric meter are analyzed by using a federated learning algorithm, the encryption and decryption speed is accelerated, and the decryption resistance is enhanced; a differential privacy protection method ...

Relationship graph method and device for extracting multiple entities and relations, equipment and medium

The invention relates to a relation graph method, device, equipment and medium for multiple entity and relation extraction, and belongs to the technical field of natural language processing named entity recognized.The method comprises the steps that firstly, a bidirectional RNN is applied to extract sequence features, then a bidirectional GCN is utilized to further extract region dependence features, and the relation between each word pair and a word entity is predicted based on the extracted word features; a complete relation weighted graph is established for each relation r, the edge in the (w1, w2) relation weighted graph is Pr (w1, w2), bidirectional GCN is adopted on each relation weighted graph, different influence degrees of different relations are considered, and comprehensive word features are aggregated; according to the method, the interaction between the entity and the relationship is considered, and the interaction between the entity and the relationship, especially the overlapped ...

Computer adaptive test method and system, electronic equipment and storage medium

The invention relates to the technical field of computer application, and provides a computer adaptive test method and system, electronic equipment and a storage medium, and the method comprises the steps: firstly determining the current capability information of a target subject for a target subject; and then inputting historical answer information and current ability information of the target subject for the target subject into the topic selection model to obtain test resources output by the topic selection model. When the basic model is trained to obtain the topic selection model, the first loss function is introduced to ensure the knowledge coverage rate of the test resources output by the topic selection model to the target subject on the whole, and the second loss function is introduced to ensure that the knowledge point correlation of the test resources output by the topic selection model on the whole is relatively small. Therefore, the credibility of the test resource can be improved ...

ADVANCED DATA COLLECTION BLOCK IDENTIFICATION

Systems and methods that allow examination of response data collected from content providers and provide for classification and routing according to the classification. The process of classification employs an unsupervised, or partially unsupervised, Machine Learning classifier model for identifying data collection responses that contains no data, mangled data, or a block, for assigning a classification correspondingly and for feeding the classification decision back to a data collection platform.

Feature map and weight selection method and accelerating device

The present disclosure provides a processing device including: a coarse-grained pruning unit configured to perform coarse-grained pruning on a weight of a neural network to obtain a pruned weight, an operation unit configured to train the neural network according to the pruned weight. The coarse-grained pruning unit is specifically configured to select M weights from the weights of the neural network through a sliding window, and when the M weights meet a preset condition, all or part of the M weights may be set to 0. The processing device can reduce the memory access while reducing the amount of computation, thereby obtaining an acceleration ratio and reducing energy consumption.

Processing method and accelerating device

The present disclosure provides a processing device including: a coarse-grained pruning unit configured to perform coarse-grained pruning on a weight of a neural network to obtain a pruned weight, an operation unit configured to train the neural network according to the pruned weight. The coarse-grained pruning unit is specifically configured to select M weights from the weights of the neural network through a sliding window, and when the M weights meet a preset condition, all or part of the M weights may be set to 0. The processing device can reduce the memory access while reducing the amount of computation, thereby obtaining an acceleration ratio and reducing energy consumption.

HIGH-TEMPERATURE DISASTER FORECAST METHOD BASED ON DIRECTED GRAPH NEURAL NETWORK

A high-temperature disaster forecast method based on a directed graph neural network is provided, and the method includes the following steps: S1, performing standardization processing on meteorological elements respectively to scale the meteorological elements into a same value range; S2, taking the meteorological elements as nodes in the graph, and describing relationships among the nodes by an adjacency matrix of graph; then learning node information by a stepwise learning strategy and continuously updating a state of the adjacency matrix; S3, training the directed graph neural network model after determining a loss function, obtaining a model satisfying requirements by adjusting a learning rate, an optimizer and regularization parameters as a forecast model, and saving the forecast model; and S4, inputting historical multivariable time series into the forecast model, changing an output stride according to demands, and thereby obtaining high-temperature disaster forecast for a future ...

Adding words to a prefix tree for improving speech recognition

An approach for improving speech recognition is provided. A processor receives a new word to add to a prefix tree. A processor determines a bonus score for a first transition from a first node to a second node in a prefix tree on condition that the first transition is included in a path of at least one transition representing the new word. A processor determines a hypothesis score for a hypothesis that corresponds to a speech sequence based on the prefix tree, where the hypothesis score adds the bonus score to an initial hypothesis score to determine the hypothesis score. In response to a determination that the hypothesis score exceeds a threshold value, a processor generates an output text sequence for the speech sequence based on the hypothesis.

Processing sequences using convolutional neural networks

Methods, systems, and apparatus, including computer programs encoded on computer storage media, for processing sequences using convolutional neural networks. One of the methods includes, for each of the time steps: providing a current sequence of audio data as input to a convolutional subnetwork, wherein the current sequence comprises the respective audio sample at each time step that precedes the time step in the output sequence, and wherein the convolutional subnetwork is configured to process the current sequence of audio data to generate an alternative representation for the time step; and providing the alternative representation for the time step as input to an output layer, wherein the output layer is configured to: process the alternative representation to generate an output that defines a score distribution over a plurality of possible audio samples for the time step.

Способ определения местоположения очага пожара с использованием многослойного рекуррентного персептрона

Настоящее изобретение относится к способам обеспечения пожарной безопасности в помещении пожароопасных объектов. Техническим результатом заявленного решения является уменьшение ошибки определения местоположения очага пожара с достоверностью до зоны очага пожара и уменьшение времени, затрачиваемого оборудованием на обнаружение очага пожара. Технический результат достигается тем, что определяют местоположение очага пожара с использованием многослойного рекуррентного персептрона, при этом для измерения факторов пожара используют многопараметрические пожарные датчики системы пожарной сигнализации, которые размещают в помещении с применением генетического алгоритма, все измерения температуры, оптической плотности, концентрации угарного газа воздушной среды и координаты расположения многопараметрических пожарных датчиков вводят в многослойный рекуррентный персептрон, при этом обучение многослойного рекуррентного персептрона проводят методом Левенберга-Маркварда и валидацию многослойного рекуррентного ...

СПОСОБ И УСТРОЙСТВО ПРОГНОЗИРОВАНИЯ ПАРАМЕТРОВ РЕЗЕРВУАРА, ОСНОВАННЫЕ НА ОГРАНИЧЕНИИ ГЕОЛОГИЧЕСКИХ ХАРАКТЕРИСТИК, И СРЕДА ХРАНЕНИЯ ДАННЫХ

Изобретение относится к области геофизики. Предлагаются основанный на геологических характеристиках способ прогнозирования параметров резервуара, устройство, компьютерная среда хранения данных и компьютерное устройство. Способ включает в себя: выбор доминирующих сейсмических атрибутов в соответствии с релевантностью между различными типами сейсмических атрибутов целевого пласта и параметрами резервуара (S100); на основе доминирующих сейсмических атрибутов, классифицирование сейсмических форм волны целевого пласта с помощью предварительно заданной модели сети классификации формы волны и в соответствии с признаками формы волны, чтобы получить результат классификации формы волны (S200); принятие результата классификации формы волны в качестве ограничения, и построение различных моделей глубокой нейронной сети, соответствующих различным геологическим характеристикам (S300); обучение различных моделей глубокой нейронной сети по сейсмическим данным и данным каротажа целевого пласта, чтобы оптимизировать ...

Soft-forgetting for connectionist temporal classification based automatic speech recognition

Material microscopic evolution prediction method and system based on auto-encoder

The invention provides a material microscopic evolution prediction method and system based on an auto-encoder. The method comprises the following steps: acquiring material microstructure image information; processing the material microstructure image information to obtain a corresponding graph representation; constructing an auto-encoder and carrying out unsupervised learning on the graph representation to obtain a dimension reduction feature vector in a material microstructure evolution process; inputting the dimension reduction feature vector into a recurrent neural network constitutive model for training to obtain an updated feature vector, and performing reconstruction by using an auto-encoder to obtain evolved material microstructure image information; and taking the evolved material microstructure image information as input, and repeatedly executing the steps, so as to predict the evolution condition of the material microstructure strain field in the deformation process. According ...

Software detection method and device and electronic equipment

The invention provides a software detection method and device and electronic equipment, and the software detection method comprises the steps: obtaining a binary file of to-be-detected software, and determining an assembly instruction sequence and a control flow diagram of the to-be-detected software according to the binary file; determining a first feature vector of the assembly instruction sequence, and determining a second feature vector of the control flow diagram; splicing the first feature vector and the second feature vector to obtain a target feature vector; and inputting the target feature vector into a pre-trained classifier for classification to obtain the category of the to-be-detected software, so that the category of the to-be-detected software can be accurately and quickly determined.

Closed-loop optimization adaptive scheduling method for semiconductor production line

The invention relates to a closed-loop optimization adaptive scheduling method for a semiconductor production line, and the method comprises the steps: carrying out the simulation generation of a production attribute set and a performance index set based on original data; a sample generation module collects a production attribute set and a performance index set to form a sample set, and performs data preprocessing; based on a dimension reduction immune algorithm, selecting an independent feature subset from a production attribute set of the first sample set of the dynamic scheduling model, and forming a second sample set with a corresponding performance index set; constructing an ELM-based performance prediction model, and training through a sample set of the performance prediction model; based on a support vector regression algorithm of particle swarm optimization, constructing a dynamic scheduling model based on performance driving, and performing training based on the second sample set ...

Method and system for tracking cattle target by unmanned aerial vehicle in pasture based on Siamese network

The invention discloses a method and a system for tracking a cattle target by an unmanned aerial vehicle in a pasture based on a Siamese network, relates to the technical field of target tracking, and is used for solving the problem of poor robustness in a tracking process due to the fact that only an initial template is used in an existing target tracking method based on the Siamese network. The method is technically characterized by comprising the following steps: for subsequent video frames during tracking, judging whether a current frame tracking result needs to update a template or not according to the change of an anchor frame ratio; if the template is to be updated, a response peak Euclidean distance detection mechanism is utilized to judge whether a similar target and a real target are close to each other in the current frame; if not, carrying out quality judgment on the current frame by utilizing a high-confidence template updating mechanism; and if the template is judged to be ...

High-precision laser angle measurement method

The invention relates to a high-precision laser angle measurement method, which comprises the following steps of: firstly, establishing a Gaussian beam model of actual laser propagation, and establishing a relation model between a beam azimuth angle resolving value and an actual beam azimuth angle according to the Gaussian beam model; secondly, on the basis of the relation model, introducing a hyperbolic tangent function, and carrying out rough estimation on the azimuth angle of the light beam to obtain a rough estimation value of the azimuth angle of the light beam; constructing a BP neural network model, configuring network parameters, and performing high-precision modeling on the rough estimation value of the azimuth angle of the light beam and the actual azimuth angle of the light beam to obtain an accurate angle measurement model; and finally, inputting the rough estimation value of the azimuth angle of the light beam into the configured BP neural network by using the accurate angle ...

Pork price prediction method based on RNN model

The invention discloses a pork price prediction method based on an RNN model, and relates to the field of price prediction, and the method comprises the steps: S101, collecting time holiday and festival data, and carrying out the arrangement of each week and holiday as data items; s102, predicting the feed price data and the substitute price data by using the trained gray model; s103, data preprocessing is carried out; s104, inputting the data processed in the step S103 into a trained pork price prediction model; s105, after the step S104, the predicted pork price is finally obtained, and the method has the beneficial effects that the RNN model is adopted, so that the data characteristics of the time holiday data, the feed price data and the substitute price data can be comprehensively utilized, especially the data preprocessing of the time holiday data; and correlation reaction can be carried out on the relation between time and festivals from the angle of spatial vectors.

Eye crack width measuring method and device and storage medium

The invention relates to an eye crack width measuring method and device and a storage medium. The method comprises the steps that a first eye bitmap in front of head-up of a user is shot and acquired; shooting to obtain a video stream in which eyelids of the user are kept immobile and eyeballs move left and right along with the tracking point in the horizontal direction; segmenting a pupil center from the first eye bitmap by adopting a neural network, obtaining a pupil center line in the vertical direction, segmenting static and dynamic intersection points from each frame of the video stream by adopting the neural network, and obtaining the contour of the eye crack through a set of the intersection points; the vertical distance of the outline of the eye crack on the pupil center line is obtained, and the width of the eye crack is obtained according to the distance. The pupil center and the pupil center line are directly recognized through neural network segmentation, the junction point ...

Image light source estimation method and device

The invention provides an image light source estimation method and device, which can be applied to the technical field of image processing. When the method is executed, firstly, a target image is input into the light source estimation model, then, the light source estimation model is utilized to obtain a light source RGB value distribution diagram, and finally, the multi-light-source condition of the target image is estimated according to the light source RGB value distribution diagram. Thus, the light source RGB value distribution diagram is obtained through the light source estimation model, the RGB values of all the pixel points of the target image are obtained, and light source estimation on the conditions of multiple light sources, shadows, light reflection and the like in the image according to the RGB values of the pixel points is achieved. Therefore, the method provided by the invention not only can be suitable for a multi-light-source scene containing the conditions of multiple ...

Classifying input examples using a comparison set

Methods, systems, and apparatus for classifying a new example using a comparison set of comparison examples. One method includes maintaining a comparison set, the comparison set including comparison examples and a respective label vector for each of the comparison examples, each label vector including a respective score for each label in a predetermined set of labels; receiving a new example; determining a respective attention weight for each comparison example by applying a neural network attention mechanism to the new example and to the comparison examples; and generating a respective label score for each label in the predetermined set of labels from, for each of the comparison examples, the respective attention weight for the comparison example and the respective label vector for the comparison example, in which the respective label score for each of the labels represents a likelihood that the label is a correct label for the new example.

Machine learning-based network model building method and apparatus

A method and an apparatus to build a machine learning based network model are described. For example, processing circuitry of an information processing apparatus obtains a data processing procedure of a first network model and a reference dataset that is generated by the first network model in the data processing procedure. The data processing procedure includes a first data processing step. Further, the processing circuitry builds a first sub-network in a second network model of a neural network type. The second network model is the machine learning based network model to be built. The first sub-network performs the first data processing step. Then, the processing circuitry performs optimization training on the first sub-network by using the reference dataset.

COMPUTE OPTIMIZATION MECHANISM FOR DEEP NEURAL NETWORKS

Embodiments provide mechanisms to facilitate compute operations for deep neural networks. One embodiment comprises a graphics processing unit comprising one or more multiprocessors, at least one of the one or more multiprocessors including a register file to store a plurality of different types of operands and a plurality of processing cores. The plurality of processing cores includes a first set of processing cores of a first type and a second set of processing cores of a second type. The first set of processing cores are associated with a first memory channel and the second set of processing cores are associated with a second memory channel.

Job-launch time reduction by node pre-configuration

Based on historic job data, a computer processor can predict a configuration of a computer node for running a future computer job. The computer processor can pre-configure the computer node based on the predicted configuration. Responsive to receiving a submission of a job, the computer processor can launch the job on the pre-configured computer node.

System and method for generating a client interaction timeline

A method for generating a client interaction timeline for interactions that occur between a client of a bank and the bank or its representatives. The method includes providing an interaction source that stores information and data for each of the interactions over all banking channels and a source of data and information that identifies each of the banks clients. The method also includes performing an application sessions process that groups the interactions that each client performs over a particular channel at a certain time frame into an interaction session having a time stamp for that time frame. The method further includes performing an application interaction process that aggregates all of the available data and information about each client and the data and information for each interaction by the client with the bank, and creating an interaction timeline for each client over all of the banking channels for a predetermined time.

SYSTEMS AND METHODS TO IDENTIFY NEURAL NETWORK BRITTLENESS BASED ON SAMPLE DATA AND SEED GENERATION

Systems and methods for determining neural network brittleness are disclosed. For example, the system may include one or more memory units storing instructions and one or more processors configured to execute the instructions to perform operations. The operations may include receiving a modeling request comprising a preliminary model and a dataset. The operations may include determining a preliminary brittleness score of the preliminary model. The operations may include identifying a reference model and determining a reference brittleness score of the reference model. The operations may include comparing the preliminary brittleness score to the reference brittleness score and generating a preferred model based on the comparison. The operations may include providing the preferred model.

Surgical endoscope employing multi-spectrum ring-illuminated surgical camera

A surgical scope employing a multi-spectrum ring illuminated surgical camera with a ring lens and a plurality of sources of light set behind the camera and positioned radially about the longitudinal axis of the lens and/or scope.

Neural programming

Methods, systems, and apparatus, including computer programs encoded on computer storage media, for neural programming. One of the methods includes processing a current neural network input using a core recurrent neural network to generate a neural network output; determining, from the neural network output, whether or not to end a currently invoked program and to return to a calling program from the set of programs; determining, from the neural network output, a next program to be called; determining, from the neural network output, contents of arguments to the next program to be called; receiving a representation of a current state of the environment; and generating a next neural network input from an embedding for the next program to be called and the representation of the current state of the environment.

Systems, methods, and computer readable media for visualization of semantic information and inference of temporal signals indicating salient associations between life science entities

Disclosed systems, methods, and computer readable media can detect an association between semantic entities and generate semantic information between entities. For example, semantic entities and associated semantic collections present in knowledge bases can be identified. A time period can be determined and divided into time slices. For each time slice, word embeddings for the identified semantic entities can be generated; a first semantic association strength between a first semantic entity input and a second semantic entity input can be determined; and a second semantic association strength between the first semantic entity input and semantic entities associated with a semantic collection that is associated with the second semantic entity can be determined. An output can be provided based on the first and second semantic association strengths.

SURGICAL ENDOSCOPE EMPLOYING MULTI-SPECTRUM RING-ILLUMINATED SURGICAL CAMERA

A surgical scope employing a multi-spectrum ring illuminated surgical camera with a ring lens and a plurality of sources of light set behind the camera and positioned radially about the longitudinal axis of the lens and/or scope.

VARIABLE PRECISION AND MIX TYPE REPRESENTATION OF MULTIPLE LAYERS IN A NETWORK

In an example, an apparatus comprises a plurality of execution units comprising at least a first type of execution unit and a second type of execution unit and logic, at least partially including hardware logic, to expose embedded cast operations in at least one of a load instruction or a store instruction; determine a target precision level for the cast operations; and load the cast operations at the target precision level. Other embodiments are also disclosed and claimed.

Neural network chip for ear-worn device

A hearing aid may include a neural network chip having tiles arranged in an array, each tile including memory, 16-128 multiplier-accumulator circuits (MACs), and routing circuitry. The memory of each tile may be configured to store a portion of elements of a matrix A comprising weights of a recurrent neural network. Each tile may be configured to receive and store elements of an activation vector X, and all tiles in a column of the array may be configured to receive the same elements of X. The plurality of tiles may be configured to perform a matrix-vector multiplication A*X by performing multiply-and-accumulate sub-operations in parallel among the plurality of tiles. The routing circuitry from the tiles in each respective row of tiles may be configured to combine results of the multiply-and-accumulate sub-operations.

Transient electromagnetic inversion method based on wavelet packet denoising and BP neural network

The invention discloses a transient electromagnetic inversion method based on wavelet packet denoising and a BP neural network. The method comprises the following steps: S1, collecting original data; s2, processing vertical magnetic field response data by using wavelet denoising; s3, standardizing input and output of a neural network data set; s4, initial parameters of the neural network are set, and a proper BP structure is selected; s5, setting initial parameters of the GA; s6, using a GA-BP algorithm to train BP, and performing BP model evaluation; s7, judging whether a final condition is met or not according to the BP model evaluation result, if so, performing the next step, and if not, continuing to perform the step S6 until the final condition is met; s8, generating a GA-BP algorithm inversion model; and S9, inverting the geoelectric model according to the test set, and obtaining an inversion result. The WPD-GA-BP joint algorithm is a new method for TEM layered geoelectric model inversion ...

System and method for adjusting varying resource restrictions

An apparatus, system, or method for processing a data sequence may involve determining availability of a computing resource; a neural network is adjusted based on the availability to set, by the neural network, restrictions on use of the computing resources during processing of the data sequence, wherein the one or more processors are configured to adjust the neural network to set the restriction includes the one or more processors configured to modify an update control included in the neural network based on a window function; and processing, by the adjusted neural network, at least a portion of the data sequence using the computing resource according to the restriction.

Coronary artery dominant type judgment method, system and device and storage medium

The embodiment of the invention provides a coronary artery dominant type judgment method, system and device and a storage medium. The method comprises the steps that a medical image of a target object is acquired; a first segmentation result and a second segmentation result are determined based on the medical image, the first segmentation result is a coronary artery segmentation result, and the second segmentation result is a heart segmentation result; coronary artery branches related to dominant type judgment are extracted from the first segmentation result, and the coronary artery branches related to dominant type judgment comprise a right coronary artery and a left convolution branch; determining a semi-center boundary plane based on the second segmentation result; and determining the dominant type of the coronary artery of the target object through a dominant type judgment model based on the right coronary artery, the left convolution branch and the half-center boundary plane.

Crop growth prediction method and device, electronic equipment and storage medium

The invention provides a crop growth prediction method and device, electronic equipment and a storage medium, and relates to the technical field of artificial intelligence information processing, and the method comprises the steps: obtaining multiple groups of growth factor data of a to-be-predicted crop; inputting the plurality of groups of growth factor data into a feature extraction layer to obtain a plurality of growth feature vectors; determining at least one vector group in the plurality of growth feature vectors, inputting the at least one vector group to a feature combination layer, and fusing the growth feature vectors in each vector group to obtain each combined feature vector; inputting the combined feature vector and the plurality of growth feature vectors into a self-attention layer, executing an attention mechanism based on the self-attention layer, respectively configuring corresponding attention weights for the combined feature vector and the plurality of growth feature ...

Method for predicting NH3 generation in organic solid waste composting process based on metabolism kinetics-GRU

The invention discloses a method for predicting NH3 generation in an organic solid waste composting process based on metabolic kinetics-GRU, relates to the technical field of environmental science, and aims to solve the problem of low prediction accuracy of NH3 generated in the organic solid waste composting process in the prior art. By means of the method, the defects that a mechanism model change mode is fixed and inaccurate are overcome, the problem that a mathematical model is insufficient in data size is solved, the NH3 output prediction model in the composting process is small in error and high in accuracy, and the final purposes of harmlessness, reduction, recycling and stabilization of organic solid waste treatment are achieved.

Customer management method based on digital twin technology and related products

The embodiment of the invention discloses a customer management method based on a digital twin technology and a related product. The method comprises the following steps: acquiring target customer information of a target customer; generating a digital twinborn body of the target customer based on the target customer information by using a digital twinborn technology; when the target customer arrives at a target physical network point, acquiring real-time customer data of the target customer; predicting a target behavior intention of the target customer according to the digital twin and the real-time customer data; determining a target management strategy parameter corresponding to the target behavior intention; and executing a target operation corresponding to the target management strategy parameter. By adopting the embodiment of the invention, the customer behavior can be pre-judged in advance, and the processing efficiency of the customer complaint event is improved.

Aero-engine turbine disk fatigue life prediction model based on artificial intelligence

The invention belongs to the technical field of thermal mechanical fatigue life prediction, and discloses an artificial intelligence-based aero-engine turbine disk fatigue life prediction model, which comprises the following steps of S1, obtaining test data; S2, performing feature screening on a test data set; s3, performing model training on the feature data set; s4, judging maximum stress prediction results of the training data set and the prediction data set; s5, outputting a maximum stress prediction result of the prediction data set; s6, searching a relationship between the feature data set and the thermal mechanical fatigue life; s7, establishing a thermal mechanical life prediction mathematical model; and S8, substituting the maximum stress of the prediction data set into the prediction mathematical model, and predicting to obtain the thermal mechanical fatigue life of the turbine disc. The method solves the problems of high test cost, long period and low life prediction precision ...

Image representation model training method and device based on image spatial position

The invention relates to the technical field of machine learning, and provides an image representation model training method and device based on an image spatial position. The method comprises the following steps: determining low-dimensional vector representation of each small image through a nonlinear coding network, and determining front and back space latent vectors of the front K small images through an autoregression network according to the low-dimensional vector representation of the front K small images; enabling the front space latent vector and the rear space latent vector of the Kth small image to pass through N-K full-connection neural networks respectively to obtain prediction vector representations of the (K + 1) th small image to the Nth small image; calculating self-supervision loss according to the low-dimensional vector representation and the prediction vector representation of the (K + 1) th small image to the Nth small image; according to the front K small images and ...

Cross-modal MRI-CT image synthesis method, system, device and medium

The invention discloses a cross-modal MRI-CT image synthesis method, system, equipment and medium, and the method comprises the steps: obtaining a two-dimensional MR slice image and a two-dimensional CT slice image, processing and expanding a training data set, inputting the training data set into a constraint consistency cyclic generation adversarial network model, and operating adversarial optimization training, and in the adversarial optimization training process, obtaining a two-dimensional MR slice image and a two-dimensional CT slice image; the game process of the discriminator and the generator is controlled through the antagonism loss item, after an optimal constraint consistency cyclic generative adversarial network model is obtained, the model is stored, the result of the optimal stage serves as output, and a final synthetic CT image is obtained; the invention further comprises a system, equipment and a medium for realizing the cross-modal MRI-CT image synthesis method, and aims ...

Micro-expression recognition method and device based on multi-task learning and global cyclic convolution

The invention discloses a micro-expression recognition method and device based on multi-task learning and global cyclic convolution, and the method comprises the steps: carrying out the preprocessing of a training data set, obtaining an amplification frame sequence with a fixed length, calculating the optical flow between adjacent frames, obtaining an optical flow sequence, and marking the coordinates of five feature points of five sense organs of a human face; constructing a convolutional neural network module I to adjust the frame size and the channel number of the amplification frame sequence; constructing a global cyclic convolution module II to extract global features; and then a multi-task learning module III including optical flow estimation, face feature point regression and three-dimensional convolutional neural network micro-expression classification prediction is constructed, and prediction is carried out. According to the method, an end-to-end deep learning framework is adopted ...

Method for training neural network in text recognition field and text recognition method

The invention discloses a method for training a neural network in the field of text recognition and a text recognition method, relates to the technical field of text recognition, and can solve the technical problem that an intention sequence cannot be recognized in current text recognition. The method for training the neural network in the field of text recognition comprises the following steps: performing word segmentation operation on a training text to convert the training text into a matrix consisting of word vectors; repeatedly executing the following steps until a loss function connected with the time sequence classification method is smaller than a preset value to obtain a trained neural network: inputting a matrix formed by word vectors into the neural network, and calculating through the neural network to obtain a feature vector; receiving the feature vector by using a connection time sequence classification method, and calculating a loss function of the connection time sequence ...

Prediction model construction method, prediction method and device based on step-by-step association weight

The invention discloses a step-by-step association weight-based prediction model construction method and device, and a prediction method and device. The method comprises the following steps of: obtaining target sample data and a plurality of associated sample data; for each associated sample data, constructing an initial prediction model corresponding to the associated sample data, and generating a first prediction model corresponding to the associated sample data; for each first prediction model, a sub-network is additionally arranged behind the first prediction model, a second prediction model is generated, the second prediction model is trained, and a third prediction model is generated; wherein the network parameters of the first prediction model are kept unchanged during training, and the network parameters of the added sub-networks are adjusted; all the third prediction models are connected through a full-connection network, a fourth prediction model is generated, the target sample ...

HARDWARE ACCELERATOR, DATA PROCESSING METHOD, SYSTEM-LEVEL CHIP, AND MEDIUM

A hardware accelerator for running an instruction set of a recurrent neural network, a data processing method, a system-level chip, and a medium are provided. The hardware accelerator is configured to process the instruction set. The instruction set includes: a data flow control instruction used for performing data flow control; a general-type computing instruction used for performing general-type computation to implement general-type computation in the recurrent neural network; a special-type computing instruction used for performing special-type computation to implement special-type computation in the recurrent neural network; an exponential shift instruction used for performing exponential shifting to implement data normalization during computation of the recurrent neural network; and a data transfer instruction used for performing data transfer to implement data transfer operations between different registers as well as data transfer operations between registers and memories during ...

Methods and systems for detection in an industrial Internet of Things data collection environment with intelligent data management for industrial processes including sensors

An apparatus, methods and systems for data collection in an industrial environment are disclosed. A monitoring system can include a data collector coupled to a plurality of sensors to collect data, a data storage structured to store a plurality of data collection management plans, a data acquisition circuit structured to interpret a plurality of detection values from the collected data, and a data analysis circuit structured to analyze the collected data and select one of the plurality of data collection management plans, wherein the selected one of the plurality of data collection management plans is selected is at least in part based on a data analysis of received data from the plurality of sensors.

SYSTEM AND METHOD FOR ADAPTING TO CHANGING CONSTRAINTS

In general, at least one example of an embodiment can involve selecting a neural network from a plurality of neural networks based on an indication of resource availability and processing data using the selected neural network in accordance with the resource availability.

Device placement optimization with reinforcement learning

A method for determining a placement for machine learning model operations across multiple hardware devices is described. The method includes receiving data specifying a machine learning model to be placed for distributed processing on multiple hardware devices; generating, from the data, a sequence of operation embeddings, each operation embedding in the sequence characterizing respective operations necessary to perform the processing of the machine learning model; processing the sequence of operation embeddings using a placement recurrent neural network in accordance with first values of a plurality network parameters of the placement recurrent neural network to generate a network output that defines a placement of the operations characterized by the operation embeddings in the sequence across the plurality of devices; and scheduling the machine learning model for processing by the multiple hardware devices by placing the operations on the multiple devices according to the placement defined ...

METHOD AND ELECTRONIC DEVICE FOR PROVIDING TEXT-RELATED IMAGE

An artificial intelligence (AI) system for simulating functions such as recognition, determination, and so forth of human brains by using a mechanical learning algorithm like deep learning, or the like, and an application thereof is provided. A method of providing a text-related image is provided. The method includes obtaining a text, determining at least one image related to the obtained text based on a degree of relatedness between a result of applying a first AI data recognition model to the obtained text and a result of applying a second AI data recognition model to a user-accessible image, and displaying the determined at least one image to a user.

COMPUTATIONAL MODEL FOR CREATING PERSONALIZED ROUTES BASED AT LEAST IN PART UPON PREDICTED TOTAL COST OF CLAIM FREQUENCY OR SEVERITY

Systems and methods are provided for providing recommendations of safe driving routes that are tailored to the driving habits of particular drivers. A machine learning model (e.g., an artificial neural network) may be trained using data indicative of insurance claim severity, road conditions, and/or vehicle telematics data associated with vehicle-related incidents, such as vehicle collisions. The machine learning model may be trained to identify road types and conditions that are predictive of claim frequency and severity. Any given driving route(s) may be provided to the trained machine learning model, and a risk value may be computed for the route(s). By further applying a personalized driver profile to the calculations of risk, personalized risk values may be computed for the route(s), and a safest route may be recommended to a driver.

Deep learning model for learning program embeddings

A system and method for using a deep learning model to learn program semantics is disclosed. The method includes receiving a plurality of execution traces of a program, each execution trace comprising a plurality of variable values. The plurality of variable values are encoded by a first recurrent neural network to generate a plurality of program states for each execution trace. A bi-directional recurrent neural network can then determine a reduced set of program states for each execution trace from the plurality of program states. The reduced set of program states are then encoded by a second recurrent neural network to generate a plurality of executions for the program. The method then includes pooling the plurality of executions to generate a program embedding and predicting semantics of the program using the program embedding.

Detection of correctness of pronunciation

There is provided automatic detection of pronunciation errors in spoken words utilizing a neural network model that is trained for a target phoneme. The target phoneme may be a phoneme in English language. The pronunciation errors may be detected in English words.

CAMERA INITIALIZATION FOR LANE DETECTION AND DISTANCE ESTIMATION USING SINGLE-VIEW GEOMETRY

Disclosed are methods, devices, and computer-readable media for detecting lanes and objects in image frames of a monocular camera. In one embodiment, a method is disclosed comprising receiving a plurality of images; identifying a horizon in the plurality of images by inputting the plurality of images into a deep learning (DL) model (either stored on a local device or via a network call); determining one or more camera parameters based on the horizon; and storing or using the camera parameters to initialize a camera.

DYNAMIC QUANTIZATION OF NEURAL NETWORKS

An apparatus for applying dynamic quantization of a neural network is described herein. The apparatus includes a scaling unit and a quantizing unit. The scaling unit is to calculate an initial desired scale factors of a plurality of inputs, weights and a bias and apply the input scale factor to a summation node. Also, the scaling unit is to determine a scale factor for a multiplication node based on the desired scale factors of the inputs and select a scale factor for an activation function and an output node. The quantizing unit is to dynamically requantize the neural network by traversing a graph of the neural network.

MULTIMODE RESERVOIR