128 results about "LTM - Long-term memory" patented technology

A minimally invasive, implantable monitor and associated method for chronically monitoring a patient's hemodynamic function based on signals sensed by one or more acoustical sensors. The monitor may be implanted subcutaneously or submuscularly in relation to the heart to allow acoustic signals generated by heart or blood motion to be received by a passive or active acoustical sensor. Circuitry for filtering and amplifying and digitizing acoustical data is included, and sampled data may be continuously or intermittently written to a looping memory buffer. ECG electrodes and associated circuitry may be included to simultaneously record ECG data. Upon a manual or automatic trigger event acoustical and ECG data may be stored in long-term memory for future uploading to an external device. The external device may present acoustical data visually and acoustically with associated ECG data to allow interpretation of both electrical and mechanical heart function.

A video recording device is capable of being mounted in a vehicle for recording images incident to the operation of the vehicle. The video recording device includes a housing mountable to a vehicle component, and at least image sensor for capturing images. A processor is coupled to the housing and is in communication with the image sensor, for processing images captured by the image sensor. An event detector is provided for detecting the existence of a designated event (such as a crash). A long term memory device is provided for storing images captured by the at least one image sensor for a time period prior to, and for a time period following the detection of a designated event by the event detector.

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.

An embodiment of the invention provides a character identifying method and system. The method includes collecting an original image of a nature scene; performing pre-treatment on the original image; performing OCR layout analysis on the original image subjected to the pre-treatment and obtaining a plurality of pixel matrixes; performing characteristic extraction on the pixel matrixes by adopting a CNN (Convolutional Neural Network) and obtaining a plurality of characteristic patterns; performing character identification on the characteristic patterns by adopting an LSTM (Long Short Term Memory) provided with an Attention Model and obtaining a character sequence, wherein a forget gate of the LSTM provided with the Attention Model is replaced with the Attention Model. According to the invention, by utilizing the LSTM algorithm provided with the Attention Model, the characteristic sequence extracted by using the CNN algorithm is identified as the corresponding character sequence, so that required text information is obtained and operation parameters are reduced. At the same time, through control of different influence on current characters by different context content, information in long term memory can be transmitted to the current characters perfectly, so that character identification accuracy is improved.

The invention provides a wireless network flow rate prediction method based on an LSTM network. The method is used for solving the technical problem of low prediction precision in the prior art. The method has the realization steps of constructing an LSTM composite network; obtaining a training set data and test set data of the LSTM composite network; initializing the parameter of an LSTM composite network; training the LSTM composite network; optimizing the trained LSTM composite network; predicting the flow rate on the future data. The long-period memory performance of the LSTM network on the flow rate data is sufficiently utilized; the contribution of the historical information on the current prediction can be automatically regulated according to the current state; the wireless networkflow rate prediction precision is improved; the method can be used in the fields of internet of vehicles, finance and the like.

The invention discloses a target tracking method based on target template updating. A target template base is established based on particle filtering principles under characteristic fusing and employing a video block multi-task means. The information of the next frame and target information in the target template base are matched and sorted by using color and edge characteristics. Template memory criteria are used to update the target template base, thereby realizing target long-term memory under different states. The target tracking method overcomes the defect that conventional video target tracking technology is likely to be influenced by external interference. Targets can be accurately positioned, and tracking errors can be effectively reduced.

The invention discloses a stock prediction method based on depth learning, which includes: obtaining the latest transaction data for the target stock and associated stock, generating a multi-dimensional feature matrix corresponding to the latest transaction data, inputting a multi-dimensional characteristic matrix corresponding to the latest transaction data into a composite neural network for processing, obtaining a prediction result of the target stock. The invention also discloses a stock prediction device based on depth learning, stock prediction apparatus and storage medium based on in-depth learning; the invention firstly utilizes the convolution neural network of the composite neural network to learn the characteristics of the transaction data of the target stock and the associatedstock, Features are input into the composite neural network of short-term and long-term memory network for processing, and the prediction of the stock price rise and fall is obtained. A stock forecastmethod based on depth learning and swarm intelligence is provided, which can accurately forecast the stock price rise and fall.

Chinese word segmentation is a process that a Chinese character string is segmented into a word sequence according to a certain specification. Since the structure of a Chinese sentence is complex and no formal delimiters are in the presence between words and even the information of a next episode needs to be combined to carry out word segmentation judgment, the accuracy of an existing Chinese word segmentation method needs to be improved. The invention discloses a Chinese word segmentation method, which comprises the following steps that: 1: inputting a Chinese text to be subjected to word segmentation into a system to serve as a sequence A; 2: transferring the sequence A to a word vector searching layer, and converting an input character into a word vector to obtain a sequence B; 3: transferring the sequence B as an input sequence to a bidirectional shot and long term memory neuron network based on an attention mechanism, and subsequently, through one layer of hidden layer, obtaining an output sequence C; 4: transferring the sequence C as the input sequence to a conditional random field decoding layer, and generating a word segmentation markup tag sequence D; and finally, converting the sequence D into a text sequence E spaced by a space.

A method for an AI time machine to accept sequential input tasks from at least one user, manage tasks, and execute tasks simultaneously or sequentially. Tasks specified by a user can be accomplished in the virtual world or in the real world and includes extracting digital data from electronic devices or manipulation of objects in the real world. The AI time machine's data structures, comprising: at least one dynamic robot to train the AI time machine; a main program with two modes: training mode and standard mode; external technologies, comprising: universal artificial intelligence programs, human level robots, psychic robots, super intelligent robots, the AI time machine, dynamic robots, a signalless technology, atom manipulators, ghost machines, a universal CPU, an autonomous prediction internet, and a 4-d computer; a videogame environment for virtual characters to do and store work; a prediction internet; a universal brain to store dynamic robot pathways or virtual character pathways, said universal brain, comprising: a real world brain, a virtual world brain, and a time machine world brain; a timeline of Earth that records predicted knowledge of Earth's past, current and future; a future United States government system; and a long-term memory. The present invention further serves as a universal AI to control at least one of the following: a machine, a hierarchical team of machines, a universal machine and a transforming machine.

The present invention provides a vision object tracking method based on hierarchical convolution. The method mainly comprises the content consisting of hierarchical convolution, correlation filters, translation estimation from rough to fine, region proposal and model updating. The method comprises the processes of: employing hierarchical features in a convolutional layer, and employing bilinear interpolation to regulate each feature map to a larger fixed dimension; performing normalization of the cycle version of input features to a soft target score generated by a Gaussian function; searchingthe maximum value of a target object on a response map; giving a related response map set; performing hierarchical deduction of each layer of target translation; calculating one confidence coefficient score of each proposal; keeping long-term memory of the target appearance; and finally, performing minimization of output errors to update an optimal filter. The vision object tracking method basedon hierarchical convolution mitigates sampling fuzziness, reduces tracking drift, reduces errors caused by reasons such as illumination change, shielding, background hybridization, sudden movement andtarget drift out of a visual field, and improves identification accuracy and robustness.

Images are captured within a camera. Upon preliminary image capturing being initiated, color balancing and auto-focus is performed and a predetermined number of images are captured for storage in a first-in-first-out buffer. This is repeated until capture is terminated. Upon a final image capture being initiated, a first image is transferred from the first-in-first-out buffer to a long-term memory within the camera.

The invention provides a video sequence loss frame prediction recovery method based on a deep neural network. The deep learning correlation theory is used, a deep convolution network is used for automatically extracting image features and the learning ability of a LSTM long-term memory network on a time sequence, a fixed number of video frame image data are used as training samples to train the network, and loss frame prediction recovery in the video sequence is performed, the intrinsic features of the video frame images and the similarity and coherence of the images between the frames are fully used to improve the prediction accuracy and efficiency, meanwhile, the video sequence loss frame prediction recovery method is high in generalization ability and has a certain social value and practical significance.

The invention belongs to the technical field of speech separation and particularly relates to a memory and attention model-based auditory selection method and device, so as to solve the problems of supervision label arrangement, speaker aliasing number uncertainty and memory unit dimension fixation in the prior art. The memory and attention model-based auditory selection method comprises steps: original speech signals are coded as a time frequency matrix, the time frequency matrix is subjected to coding and transformation, the time frequency matrix is converted to speech vectors, a long term memory unit is used to store speakers and corresponding speech vectors, the speech vector of a target speaker is acquired, and target speech is separated from the original speech signals through an attention selection model. The method provided in the invention does not need to fix or designate the number of speakers but can separate the target speech from the original speech signals.

Applicable to the setting method of the compound characteristic impedance of the series capacitor compensation line, ①Construct a conventional ground distance relay and a phase-to-phase distance relay; ②Construct a reactance relay used in conjunction with a distance relay; ③Based on the above-mentioned conventional relay and impedance relay, set different Time memory characteristics, forming two impedance relays with different memory times, that is, zero-sequence reactance relays and reactance relays, and the memory times are t and T respectively; If the relays operate at the same time, it is considered as a positive fault, and the distance protection is judged as a positive direction; if the short-memory impedance relay acts before the long-memory impedance relay, the blocking logic starts, and the distance protection is judged as a reverse direction.

The invention relates to a cationic phospholipid-polymer hybridized nanoparticle vaccine adjuvant of a common-carrier antigen, MPLA (Monophosphoryl Lipid A) and IMQ (Imiquimod) as well as a preparation method and application thereof. The vaccine adjuvant is characterized in that the IMQ as a TLR7 agonist is loaded on a hydrophobic core; the MPLA as a TLR4 agonist is loaded in a phopholipid layer;cationic phospholipid DOTAP (1,2-dioleoy-3-trimethylammonium-propane) in the phopholipid layer is used for adsorbing an antigen; the antigen is protected through hybridized nanoparticles, and the ingestion of the antigen by dendritic cells is improved; immune response after antigen stimulation is improved remarkably through the TLR agonist, and cross-presentation of the antigen is improved remarkably. The hybridized nanoparticles as the vaccine adjuvant can load the antigen and different types of TLR agonists simultaneously, can deliver the antigen through a plurality of immune paths, and promotes the DC activation and maturation. The cross-presentation level is raised, a strong and powerful T-cell killing effect is achieved, cell factor secretion is induced, a long-term memory T-cell reaction is generated, and higher prevention capability for tumors is achieved.

The invention relates to a vehicle trajectory network interactive predicting method for the movement states of a plurality of vehicles, and belongs to the technical field of intelligent networking vehicle environment awareness. The method provided by the invention comprises the following steps: determining the relative position relationship of a plurality of networked vehicles; extracting the characteristics of the driving state of each vehicle, respectively inputting the characteristics of the driving states of the vehicles into short and long term memory units, and connecting and sharing hidden states of adjacent vehicles through a radial network so as to construct structured short and long time memory units and realize the modelling of an interaction relationship among the vehicles; then, establishing a multilayer encoder-decoder network to predict the future driving trajectories of the vehicles by utilizing the structured short and long time memory units; finally, transmitting thepredicted state obtained by the network to a decision making module to perform autonomous decision-making and path planning. According to the vehicle trajectory network interactive predicting method for the movement states of the vehicles, the driving states of the networked vehicles are shared in a hidden state layer by adopting the structured short and long time memory units so as to realize themodelling of the interaction relationship; by the method, synchronous long-time span prediction can be performed on the vehicles in a complex traffic environment to achieve high prediction accuracy.

The invention provides a method which uses Hurst parameter values to represent self-similarity presented by network traffic, and analyzes statistic characteristics of the network traffic in big time scale in combination with the R/S series analysis method. Since the R/S series analysis method can better describe fractal characters and long-time memory process of the network traffic, and the FARIMA model which is used at later stage can also describe long-range dependence and short-range dependence of the network traffic, according to the invention, the method can conduct long-time prediction on the network traffic in a precise manner, has better astringency, and is very significant in terms of increasing network properties and service quality.

The invention discloses a text sentence segmentation method and system. The method comprises following steps: pre-collecting a small amount of textual data and corresponding speech data, constructinga long-term memory segmentation model based on text segmentation features and acoustic segmentation features; when the text is segmented, obtaining the text of the sentence to be segmented and corresponding speech data; extracting text segmentation features and the acoustic segmentation features according to the to-be-segmented text and the speech data corresponding to the to-be-segmented text, respectively; according to the extracted text segmentation features, acoustic segmentation features, and the long-term memory segmentation model, segmenting the to-be-segmented text. The invention can effectively improve the accuracy of text segmentation.

The invention relates to the field of food biotechnology, and in particular to a germinant fagopyrum tararicum medicated-food composite powder rich in gamma-aminobutyric acid and a preparation process of the composite powder. The germinant fagopyrum tararicum medicated-food composite powder is prepared from the following components in parts by weight: 40-60 parts of germinant fagopyrum tararicum powder, 20-35 parts of radix puerariae powder, 10-20 parts of medicated-food composite powder, 10-15 parts of dextrin and 0.04-0.06 part of a sweetening agent. The preparation process comprises the steps that the germinant fagopyrum tararicum powder, the radix puerariae powder, the medicated-food composite powder, the dextrin and the sweetening agent are uniformly mixed in a ratio by weight, the mixture is roasted until an obvious fragrance is sent out, and finally, the product is packaged in vacuum or by charging nitrogen. The content of GABA (gamma-aminobutyric acid) in the germinant fagopyrum tararicum medicated-food composite powder prepared by the preparation process disclosed by the invention can reach 20-40 mg/100 g, the total flavonoid content is up to10-15 mg/100 g, and the nutritional ingredients in the medicated-food composite powder are greatly supplemented; the germinant fagopyrum tararicum medicated-food composite powder has the functions of reducing blood pressure, blood fat and blood sugar, improving brain functions, stabilizing nerves, promoting long-term memory and enhancing body immunity.

Disclosed herein is technology for learning a foreign language based on the stimulation of long-term memory. The foreign language learning method based on the stimulation of long-term memory comprises associative learning which is based on image content and keyword information related thereto, learning which allows a learner to listen to and speak foreign language sentences that comprise keyword information and are related to the image content, learning which allows the learner to repeatedly listen to and speak words constituting a foreign language sentence in such a way as to sequentially add words one by one in a sequence that the words appear in the sentence, and learning which allows the learner to naturally speak sentences with which a specific image is associated when viewing the image, so that the learner habitually memorizes each foreign language sentence or remembers it as one episode, thus providing efficient learning.

Various aspects for managing input/output (I/O) operations for data objects (e.g., large data objects (LOBs)), in a storage system are provided. For instance, a method may include receiving an I/O request for executing an I/O operation on a data object, determining the size of the data object, and determining the data object to be a non-conforming data object relative to a storage threshold of a cache memory device. The method may further include dividing the non-conforming data object into multiple data chunks, each of the data chunks having a size that is less than the storage threshold, moving one or more data chunks stored in one or more memory spaces of the cache memory device to a long-term memory device to free up the one or more memory spaces, and storing one or more of the data chunks of the non-conforming data object in the freed-up memory space(s).

The invention discloses a behavior modeling method of a power amplifier based on a clock recurrent neural network. The problems that an ordinary neural network model is iterated many times, and the long-term memory effect performance is poor are solved. By means of the characteristics that output of a recurrent neural network is not only related to real time input but also related to historical input, the memory effect of the power amplifier is described. On the basis, the weight of an ordinary recurrent neural network hidden layer is divided into multiple modules, each module has its own cycle, the weight of the module is only updated in its own cycle, and the number of weight updates is reduced to accelerate training of the neural network model. Accordingly, the non-linear features and memory effect of the power amplifier can be well described, and the high precision is achieved.

The invention belongs to the technical field of road design, and relates to a method for confirming set positions of speed limit signs and size of speed limit during road construction, and the method comprises the following steps of: (1) designing a working memory structure of a driver SOAR intellectual body; (2) designing an initial long-term memory rule, and building a long-term memory rule base of the driver SOAR intellectual body; (3) building a decision period of the driver SOAR intellectual body; and (4) setting the set positions of the different speed limit signs, the size of speed limit, and different road loading coefficients so as to perform simulation, judging traffic conflict and severity degree according to overlapping degree of road cells which are occupied by a current vehicle and adjoining vehicles, selecting a simulation condition with the lower traffic conflict and severity degree, and obtaining the set positions of the speed limit signs and the size of the speed limit within a proper construction operation area under different traffic conditions. The method can be used for more accurately reflecting the influence of the set positions of the speed limit signs and the size of the speed limit on the traffic capacity during road construction, thus providing the method for setting the speed limit signs during road construction.

A Simple Device for creating exact copies of computer long-term memory devices such as hard drives and compact flash memory. Our current invention is a stand-alone device. A user connects a long-term memory device he desires to make a copy of (source) to our device. The user also connects a long-term memory device to receive this copy (destination). Our device contains logic and circuitry, which perform operations on both the source and destination device to make an exact copy of the Source Data to the Destination Device, while protecting the Source device from any changes. Our device has a very simple user interface. This simplified user interface makes it difficult and unlikely to use our device incorrectly.