612 results about "Short-term memory" patented technology

<heading lvl="0">Abstract of Disclosure< / heading> Apparatus utilizing electrical activity of the brain to control a series of low-stimuli educational exercises displayed on a computer monitor to increase the following educational components: time on-task, visual tracking, short-term memory, visual discriminatory processing, auditory discriminatory processing, and focus. The exercises are governed by real-time analysis of the focus and processing states of the user. Specific relative exercise performance data are collected and recorded from the use of each of the educational components to demonstrate improvement over time of the user in each of the sited educational components.

A method and apparatus for natural language interpretation are described. The invention includes a schema and apparatus for storing, in digital, analog, or other machine-readable format, a network of propositions formed of a plurality of text and / or non-text objects, and the steps of retrieving a string of input text, and locating all associated propositions in the network for each word in the input string. Embodiments of the invention also include optimization steps for locating said propositions, and specialized structures for storing them in a ready access storage area simulating human short-term memory. The schema and steps may also include structures and processes for obtaining and adjusting the weights of said propositions to determine posterior probabilities representing the intended meaning. Embodiments of the invention also include an apparatus designed to apply an automated interpretation algorithm to automated voice response systems and portable knowledge appliance devices.

The technology disclosed proposes using a combination of computationally cheap, less-accurate bag of words (BoW) model and computationally expensive, more-accurate long short-term memory (LSTM) model to perform natural processing tasks such as sentiment analysis. The use of cheap, less-accurate BoW model is referred to herein as “skimming”. The use of expensive, more-accurate LSTM model is referred to herein as “reading”. The technology disclosed presents a probability-based guider (PBG). PBG combines the use of BoW model and the LSTM model. PBG uses a probability thresholding strategy to determine, based on the results of the BoW model, whether to invoke the LSTM model for reliably classifying a sentence as positive or negative. The technology disclosed also presents a deep neural network-based decision network (DDN) that is trained to learn the relationship between the BoW model and the LSTM model and to invoke only one of the two models.

Systems, methods and articles of manufacture for modeling a joint language-visual space. A textual query to be evaluated relative to a video library is received from a requesting entity. The video library contains a plurality of instances of video content. One or more instances of video content from the video library that correspond to the textual query are determined, by analyzing the textual query using a data model that includes a soft-attention neural network module that is jointly trained with a language Long Short-term Memory (LSTM) neural network module and a video LSTM neural network module. At least an indication of the one or more instances of video content is returned to the requesting entity.

A system and method for predicting driver behavior and generating control and / or warning signals comprises: one or more sensors; a database; a driver predictor for generating warning and control signals; a notification unit; and a vehicle control unit. More specifically, the driver predictor includes a transfer module, a hierarchical temporal memory and a prediction retrieval module. The transfer module is used to receive data and translate it into a form so that it can be stored in the hierarchical temporal memory. The hierarchical temporal memory receives data input and stores it such that data items are translated into short-term memory items which are in turn translated into intermediate-term data which in turn are finally translated into long-term data. The translation of data from one level of the memory to another is performed according to the present invention such that the data at each level of memory is reflective of and predictive of a particular driver behavior. Once the data is stored in the memory, the prediction retrieval module accesses the hierarchical temporal memory to generate warning signals to alert the driver of potentially dangerous conditions and / or control systems of the vehicle to prevent or avoid collisions. The present invention also includes a variety of method including a method for storing driving data in a hierarchical temporal memory, and a method for using a hierarchical temporal memory to generate collision avoidance signals.

The present invention discloses a deep learning method-based short-term traffic flow prediction method. The influence of the traffic flow rate change of the neighbor points of a prediction point, the time characteristic of the prediction point and the influence of the periodic characteristic of the prediction point on the traffic flow rate of the prediction point are considered simultaneously. According to the deep learning method-based short-term traffic flow prediction method of the invention, a convolutional neural network and a long and short-term memory (LSTM) recurrent neural network are combined to construct a Conv-LSTM deep neural network model; a two-way LSTM model is used to analyze the traffic flow historical data of the point and extract the periodic characteristic of the point; and a traffic flow trend and a periodic characteristic which are obtained through analysis are fused, so that the prediction of traffic flow can be realized. With the method of the invention adopted, the defect of the incapability of an existing method to make full use of time and space characteristics can be eliminated, the time and space characteristics of the traffic flow are fully extracted, and the periodic characteristic of the data of the traffic flow is fused with the time and space characteristics, and therefore, the accuracy of short-term traffic flow prediction results can be improved.

The invention discloses a method and a system for enhancing features by the aid of bidirectional long-term and short-term memory recurrent neural networks. The method includes steps of training inputted features with noise and corresponding features without noise by the aid of bidirectional long-term and short-term memory recurrent neural network models; enhancing the features by the aid of the bidirectional long-term and short-term memory recurrent neural network models. The method and the system have the advantages that the long-term and short-term memory recurrent neural network models can be built in two directions by the aid of the inputted features, and accordingly models of contexts of current frames can be effectively built; long-term and short-term memory cells and certain control variables are introduced into the neural networks, and accordingly the models can be built in the dependence on the long contexts; excellent enhancement effects can be realized by the models owing to long-term dependence when the features are enhanced by the aid of the models, and accordingly the performance of voice recognition systems and audio event classification systems can be improved.

The invention claims to protect a character recognition system and method based on the combination of a neural network and an attention mechanism, the system comprising: a convolution neural network feature extraction module, which is used for spatial feature of character image; The spatial features extracted by the convolution neural network are input to the bi-directional long-short memory network module, and the bi-directional long-short memory network can extract the sequence features of characters. The extracted feature vectors are semantically encoded, and then the attention weights of feature vectors are assigned through the attention mechanism, so that the attention is focused on the feature vectors with higher weights. In the decoding part of the model, the features extracted fromattention and the prediction information of the previous time are used as the inputs of the nested long-short memory network. The purpose of using the long-short memory network is to keep the temporal characteristics of the eigenvectors and make the attention points of the model constantly change with time. In the decoding part, the features extracted from attention and the prediction informationof the previous time are used as the inputs of the nested long-short memory network. The invention can more accurately detect the text area in the natural scene, and has good detection effect on thesmall target text and the text with small tilt angle.

The invention relates to a text key information identification method, an electronic apparatus and a readable storage medium. The method comprises the steps of, after a to-be-identified text is received, performing segmentation processing on the received to-be-identified text by using a predetermined word segmentation model to obtain segmented words of the to-be-identified text, wherein the predetermined word segmentation model is a long and short term memory recurrent neural network model obtained by training a preset quantity of sample statements labeled by adopting a sequence labeling method; and based on word frequencies, positions and word spans of the segmented words in the to-be-identified text, and according to a preset scoring formula, performing calculation to obtain scores of the segmented words, sorting the segmented words in the to-be-identified text according to a sequence of the scores from high to low, extracting the segmented words sorted in front to serve as keywords,and according to the extracted keywords, obtaining key information of the to-be-identified text. A user can be enabled to quickly and accurately obtain the key information in the to-be-identified text.

The invention discloses a prosodic hierarchy model training method for text-to-speech and a text-to-speech method and a text-to-speech device by a prosodic hierarchy model. The training method includes: training massive non-tagged corpus data to obtain character vectors of individual characters; obtaining textual features and tagging corresponding to the trained data according to the character vectors and prosodic tagging data; based on a deep neural network and a bidirectional LSTM (long-short term memory) neural network, training the prosodic hierarchy model according to the textual features and tagging of the trained data. The training method has the advantages that when a character granularity based dictionary in the training method is compared with a traditional word granularity based dictionary, entry scale is decreased effectively, requirements of the model and resource files on computing resources and memory spaces are lowered, and usability of a prosodic prediction model in embedded intelligent devices is guaranteed while performance of the prosodic prediction model is improved.

A sliding side menu GUI interface for computerized devices that shows indicia of updated content on at least some of its various menu items. These indicia can include icons representing graphics or thumbnail images representing individuals that caused the updated content, type of updated content, amount of updated content, or other useful information. This enables the user, while operating an original application screen, while still retaining the original screen in short term human memory, to quickly call up the sliding side menu, look for indicia of updated content or interesting new content, and if there is no such content return to the original screen without significant human short term memory loss of the original screen. The method may be implemented by server software, device software, or combination of server and device software. The method is well suited for small touchscreen devices such as smartphones and tablets.

The present invention provides an end point detection method and system of a voice identification system, wherein the method comprises the steps of training an acoustics identification model based on a long / short term memory neural network; primarily identifying a voice end point of a to-be-identified voice signal by a preset voice end point detection algorithm; extracting the voice characteristic information of the to-be-identified voice signal frame by frame, and inputting the voice characteristic information in the acoustics identification model to enable the acoustics identification model to generate an acoustics identification result of the to-be-identified voice signal according to the voice characteristic information; adjusting the primarily identified voice end point according to the acoustics identification result. The end point detection method of the voice identification system of the embodiment of the present invention provides an end point detection mode of adjusting the primarily identified voice end point by the acoustics identification result, thereby positioning the voice end point of the to-be-identified voice signal accurately, improving the voice end point detection accuracy, and further being able to improve the voice identification accuracy, and improving the performance of the voice identification system.

A computerized method for collecting suspected data of interest from a computer comprises searching the computer's shot-term memory to locate at least one target memory range containing the suspected data of interest, and copying the suspected data of interest within the target memory range to an alternate data storage location in a manner which avoids writing the suspected data to the computer's long-term memory. Alternatively, the suspected data of interest can be copied to a previously unused data storage location while preserving integrity of non-volatile memory resources. A computer-readable medium and a system for collecting target forensics data are also provided.

The invention provides a continuous voice recognition method based on a deep long and short term memory recurrent neural network. According to the method, a noisy voice signal and an original pure voice signal are used as training samples, two deep long and short term memory recurrent neural network modules with the same structure are established, the difference between each deep long and short term memory layer of one module and the corresponding deep long and short term memory layer of the other module is obtained through cross entropy calculation, a cross entropy parameter is updated through a linear circulation projection layer, and a deep long and short term memory recurrent neural network acoustic model robust to environmental noise is finally obtained. By the adoption of the method, by establishing the deep long and short term memory recurrent neural network acoustic model, the voice recognition rate of the noisy voice signal is improved, the problem that because the scale of deep neutral network parameters is large, most of calculation work needs to be completed on a GPU is avoided, and the method has the advantages that the calculation complexity is low, and the convergence rate is high. The continuous voice recognition method based on the deep long and short term memory recurrent neural network can be widely applied to the multiple machine learning fields, such as speaker recognition, key word recognition and human-machine interaction, involving voice recognition.

The invention relates to a remote sensing image natural language generation method based on an attention mechanism and deep learning. The method comprises the steps of first, preprocessing a remote sensing image and corresponding natural language description; second, inputting the denoised remote sensing image into an intensive positioning convolution neural network (IPCNN); third, inputting region blocks obtained in the second step into a reassignment long-short term memory (RLSTM) network, getting into a weight assignment layer of the RLSTM network, respectively solving the weight of each region by using a multilayer network function, and finally realizing overall output of the natural language description through a deep output layer of the RLSTM network; fourth, inputting the natural language description generated in the third step into a remote sensing image language description scoring model to obtain a score of each sentence; and fifth, inputting a target position, a category label and a natural language description score into a database to wait for calling.

The invention belongs to the technical field of an intelligent memory method, and particularly relates to a word memory intelligent learning method and a system thereof. According to the memory ability of different learners, times spent on the memory of words in each day and the degree of difficulty of memorizing different words, the forgetting models of each word in different periods (transient memory, short-term memory and long-term memory) are calculated, then according to a forgetting critical point, the optimal review time of each word is accurately calculated, thus new word amount of each memory and old word amount of review are planned, and the overall learning progress can be predicated. The learning system established according to the above method is accurate and rapid, a lot of times and energy are saved for the learner, thus the efficiency of word memory is raised, and the system is especially suitable for memorizing a large amount of words. This system is applied to a computer or various smart mobile devices.

The invention relates to a neural network text classification method based on a multi-knowledge map. The method comprises the following steps: inputting a text in a training set into a long-term and short-term memory network to obtain a context vector of the text; extracting entities from each text in the training set, and performing entity matching in the knowledge map; calculating the attentionweights of matched entities and relationships in the knowledge map under context vector to obtain the overall entity vector, the overall relationship vector and the fact triple vector of the text; computing the fact triple vectors under different knowledge maps, calculating the attention weights of these fact triples, obtaining the text representation vectors and inputting them to the full connection layer of the neural network, and using a classifier to calculate the probability of each text belonging to different categories to train the network; using the trained deep neural network model to predict the category of the text to be predicted. This method improves the model's understanding of text semantics and can classify text content more reliably, accurately and robustly.

A processor architecture for a learning machine is presented which uses a massive array of processing elements having local, recurrent connections to form global associations between functions defined on manifolds. Associations between these functions provide the basis for learning cause-and-effect relationships involving vision, audition, tactile sensation and kinetic motion. Two arbitrary images hold each other in place in a manifold association processor and form the basis of short-term memory.

The invention relates to a universal single channel real-time noise-reduction method. The universal single channel real-time noise-reduction method comprises the following steps that noisy voice of an electronic format is received and comprises voice and non-human-voice interference noise; a short-time Fourier magnitude spectrum is extracted frame by frame from the received voice to serve as an acoustic characteristic; a specific value film is generated frame by frame through a deep recurrent neural network with long-and-short-term memories; the magnitude spectrum of the noisy voice is sheltered through the generated specific value film; and the sheltered magnitude spectrum and an original phase of the noisy voice are used, and through inverse Fourier transform, a voice waveform is synthesized again. According to the universal single channel real-time noise-reduction method, voice noise reduction is conducted through a supervised learning method, and the ideal specific value film is estimated through the recurrent neural network with the long-and-short-term memories; and the recurrent neural network provided by the invention is trained through the large amount of noisy voice, various realistic acoustic scenes and microphone impulse responses are included, and finally universal voice noise reduction independent of background noises, speakers and transmission channels is achieved.

The present disclosure pertains to a paraphrase generation system. The system comprises one or more hardware processors and / or other components. The system is configured to obtain a training corpus. The training corpus comprises language and known paraphrases of the language. The system is configured to generate, based on the training corpus, a word-level attention-based model and a character-level attention-based model. The system is configured to provide one or more candidate paraphrases of a natural language input based on both the word-level and character-level attention-based models. The word-level attention-based model is a word-level bidirectional long short term memory (LSTM) network and the character-level attention-based model is a character-level bidirectional LSTM network. The word-level and character level LSTM networks are generated based on words and characters in the training corpus. In some embodiments, the LSTM networks are stacked residual LSTM networks comprising residual connections between stacked layers of a given LSTM network.

The invention discloses a wind power short-term prediction method based on a long-term and short-term memory network, comprising a long-term and short-term memory neural network training algorithm, ashort-term wind power prediction error distribution algorithm and a wind turbine generator power short-term prediction model design. A long-term and short-term memory network algorithm (LSTM)-based wind pow prediction model is established based on the depth learning network, and the Gaussian mixture model (GMM) is used to analyze the error distribution characteristics of the short-term wind powerprediction. The invention can obtain different confidence intervals of two units through the GMM model. It is proved that LSTM method has higher precision and faster convergence rate, and GMM method has practical application value for wind power dispatching.

Disclosed is a deep long-term and short-term memory recurrent neural network acoustic model establishing method based on selective attention principles. According to the deep long-term and short-term memory recurrent neural network acoustic model establishing method based on the selective attention principles, attention gate units are added inside a deep long-term and short-term memory recurrent neural network acoustic model to represent instantaneous function change of auditory cortex neurons; the gate units are different in other gate units in that the other gate units are in one-to-one correspondence with time series, while the attention gate units represent short-term plasticity effects and accordingly have intervals in the time series; through the neural network acoustic model obtained by training mass voice data containing Cross-talk noise, robustness feature extraction of the Cross-talk noise and establishment of robust acoustic models can be achieved; the aim of improving the robustness of the acoustic models can be achieve by inhibiting influence of non-target flow on feature extraction. The deep long-term and short-term memory recurrent neural network acoustic model establishing method based on the selective attention principles can be widely applied to multiple voice recognition-related machine learning fields of speaker recognition, keyword recognition, man-machine interaction and the like.

An identification system is provided for identifying a plurality of randomly aligned object-based transponders (18, 20, 22) passing through a predetermined reading volume. A reader (10) includes multiple transmitting and receiving antenna arrays (12, 14, 16) which are arranged sequentially to transmit interrogating and energizing fields (29) into the reading volume, with each energizing field having a different polarisation. Each transponder (12, 14, 16) is arranged to be powered by at least one of the interrogating and energizing fields, and includes a short term memory module (30) which is arranged to record the successful transmission or otherwise of the identification signal of the transponder (20) and to retain the record in the absence of a power supply to the transponder. The short term memory module has a retention time longer than any predicted interruption of power supply to the transponder via one of the energizing fields within the reading volume. The minimum memory retention time is typically the time period taken for the sequential transmission of all but one of the interrogating and energizing fields (29) during an interrogation cycle.

The invention discloses an intelligent man-machine conversation system in a closed domain. The system comprises a first modeling module, a second modeling module and a third modeling module, wherein the first modeling module is used for constructing a multi-feature fusion deep intention recognition model on the basis of a bidirectional long / short-term memory network and a convolutional neural network; the second modeling module is used for constructing an MC-BLST-MSCNN-based conversation state tracking model by adoption of a man-machine conversation system current state input and context statement combined modeling manner; and the third modeling module is used for constructing a Bi-LSTM matching model on the basis of an extraterritorial recover mechanism of a displacement attention mechanism, and inputting recognized user intentions and user slot values into a displacement network to carry out weight distribution the attention mechanism, so as to realize conversation state encoding andconversation control matching. The system has higher intention recognition accuracy, conversation state tracking correctness and conversation control stability, so as to enhance the cognition intelligence capability of man-machine systems.

The invention provides a social emotion classification method based on multimodal fusion, which relates to information in the form of audio, visual and text. Most of affective computing research onlyextracts affective information by analyzing single-mode information, ignoring the relationship between information sources. The present invention proposes a 3D CNN ConvLSTM (3D CNN ConvLSTM) model forvideo information, and establishes spatio-temporal information for emotion recognition tasks through a cascade combination of a three-dimensional convolution neural network (C3D) and a convolution long-short-term memory recurrent neural network (ConvLSTM). For text messages, use CNN-RNN hybrid model is used to classify text emotion. Heterogeneous fusion of vision, audio and text is performed by decision-level fusion. The deep space-time feature learned by the invention effectively simulates the visual appearance and the motion information, and after fusing the text and the audio information,the accuracy of the emotion analysis is effectively improved.

A processor architecture for a learning machine is presented which uses a massive array of processing elements having local, recurrent connections to form global associations between functions defined on manifolds. Associations between these functions provide the basis for learning cause-and-effect relationships involving vision, audition, tactile sensation and kinetic motion. Two arbitrary images hold each other in place in a manifold association processor and form the basis of short-term memory.

The embodiment of the invention discloses a self-adaptive Chinese word segmentation method based on embedded representation and belongs to the field of information processing. The method is characterized in that an embedded representation layer of a character is shared by a word segmentation network and a character language model. As for embedded representation of the character, on the one hand, hidden multi-granularity local features of a to-be-segmented text is obtained by means of the word segmentation network based on convolutional neural network; then label probability of the character is obtained through a forward network layer; finally, label inference is used to obtain the optimum segmentation result in the sentence level; on the other hand, an unlabelled text is randomly extracted, a character next to the character is predicted by means of a character language model based on a long- and short-term memory unit (LSTM) recurrent neural network and the word segmentation network is constrained. By modeling a character co-representing relationship in texts in different fields by means of the character language model and transferring information to the word segmentation network by means of embedded representation, the field transfer ability of word segmentation is enhanced, and the method has very huge practical value.

The invention discloses an abnormity detection method for satellite telemetry data through an adversarial network autoencoder, and the method comprises the steps: breaking the limitation of a traditional empirical model, and employing a pure data driving model; on the basis of a variational autoencoder, introducing a confrontation network idea, using a bidirectional LSTM (Long Short Term Memory) (Long-short term memory network) as a discriminator, and judging whether satellite telemetry data is abnormal or not by using errors of reconstructed data and original data; aiming at the redundancy problem of a satellite sensor, the conventional situation is broken through, and a Markov distance is used for measuring a reconstruction error. In combination with periodicity of satellite orbit operation, a dynamic threshold determination method based on a periodic time window is provided. The method has the advantages that pure data driving is adopted, expert experience is not needed, and the method can be suitable for various occasions; By combining the respective advantages of the variational auto-encoder and the generative adversarial network, the proposed network has the characteristics of high training speed and relatively easy convergence; eliminating redundant data influence between satellite telemetry data by adopting a Mahalanobis distance. According to the periodicity of the satellite, the dynamic threshold method based on the periodic time window is provided, and the misjudgment rate is reduced.

The invention provides a statement classification method and system, electronic equipment, a refrigerator and a storage medium. The statement classification method comprises the following steps of: obtaining a word vector corresponding to each word in a target statement, wherein the word vector comprises a word meaning vector and a part-of-speech vector; inputting the word vector corresponding to each word in the target statement into a long / short-term memory LSTM model so as to obtain a state matrix of the target statement, wherein the state matrix comprises initial semantic information and word order information corresponding to the target statement; inputting the state matrix into a convolutional neural network CNN model so as to obtain a characteristic quantity corresponding to the target statement; and classifying the target statement according to the characteristic quantity of the target statement. According to the method, the correctness of statement classification and statement type recognition can be improved and the user experience can be enhanced.