39 results about "Feature discovery" patented technology

Embodiments are directed to a method of performing autonomous learning for updating input features used for an artificial intelligence model, the method comprising receiving updated data of an information space that includes a graph of nodes having a defined topology, the updated data including historical data of requests to the artificial intelligence model and output results associated with the requests, wherein different categories of input data corresponds to different input nodes of the graph. The method may further comprise updating edge connections between the nodes of the graph by performing path optimizations that each use a set of agents to explore the information space over cycles to reduce a cost function, each connection including a strength value, wherein during each path optimization, path information is shared between the rest of agents at each cycle for determining a next position value for each of the set of agents in the graph.

Methods, apparatus and computer readable code for determining whether a potential relay device is a relay device are provided herein. In some embodiments, first and second information elements are received from a potential relay device, which is an original source of the second information element. In order to determine whether the potential relay device is a relay device, it is determined whether a feature of an original source of the first information element and a feature of the potential relay device are features unlikely to relate to a single device, wherein a positive result of the determining is indicative that the potential relay device is a relay device. In an exemplary embodiment, a disclosed system includes an information element receiver and a feature incompatibility analyzer. Optionally, the disclosed system includes a feature discovery module, a parameter obtainer and a feature database.

The invention relates to the technical field of information recommendation, in particular to an information recommendation method based on potential communities. The method includes: digging out user interest models from history records of a user via feature discovery technology, discovering the potential communities according to the user interest models and digging out hierarchical relationships between the communities; extracting possible interest collections from objects to be recommended through feature extraction, and fast positioning the communities to be recommended according to the features of the hierarchical relationships between the communities; calculating similarity between the communities to be recommended and the interest collections of the objects to be recommended to determine whether recommendation conditions are satisfied. The information recommendation method based on the potential communities is capable of realizing flexible batch pushing of information, enhancing efficiency greatly compared with traditional point-to-point recommendation technique, and enhancing novelty of information recommendation by integrating collaborative filtering technique.

The invention discloses an Alzheimer's disease brain network feature extraction method based on a continuous coherence technology, and belongs to the technical field of brain network analysis. According to the invention, through data preprocessing, brain network division, brain network construction, continuous coherent brain network filtering flow construction, continuous interval data statisticalanalysis and Alzheimer's disease brain network feature extraction, brain network feature extraction of a patient is realized. By constructing the multi-scale brain network with the variable thresholdvalue, the threshold value selection problem in a graph theory method is avoided, wherein the continuous feature discovery mechanism of the network complex flow can effectively reduce the calculationburden, and is an effective brain network analysis technology. The continuous homology theory is applied to the field of brain analysis of patients suffering from Alzheimer's disease, the brain mechanism is researched, the brain network connection characteristics of diseases are discovered, and the method is an innovative research idea and has important significance for early diagnosis, drug development and diagnosis and treatment scheme formulation of Alzheimer's disease.

The invention discloses a big-data-oriented privacy information release exposure chain discovery method which is based on the theory of ontology mapping. The method comprises the following steps: (1) measuring and calculating the definition similarity simd, attribute similarity simT and structure similarity sims of privacy data; (2) measuring and calculating the exposure cost of the privacy data through an exposure vector of the privacy data; (3) acquiring the characteristic attribute of a privacy data information exposure chain through measurement and calculation of the definition similarity and exposure cost of the privacy data; and (4) discovering a privacy exposure chain in data released by a user according to the characteristic of the exposure chain, thus protecting the safety in the process of user privacy data information release. Most of the existing privacy protection technologies use a way of encryption or anonymity. However, privacy data must be plaintext when participating in cloud computing. The method is a privacy data release method in the computing process, and can effectively prevent user's privacy data from being leaked in the process of multi-party service computing.

Methods and systems for detecting defects on a wafer are provided. One method includes acquiring output for a wafer generated by an inspection system. Different dies are printed on the wafer with different process conditions. The different process conditions correspond to different failure modes for the wafer. The method also includes comparing the output generated for a first of the different dies printed with the different process conditions corresponding to a first of the different failure modes with the output generated for a second of the different dies printed with the different process conditions corresponding to a second of the different failure modes opposite to the first of the different failure modes. In addition, the method includes detecting defects on the wafer based on results of the comparing step.

Systems, methods, and computer-readable storage devices are disclosed for improved audio feature discovery using a neural network. One method including: receiving a trained neural network model, the trained neural network configured to output an audio feature classification of audio data; deconstructing the trained neural network model to generate at least one saliency map, the at least one saliency map providing a successful classification of the audio feature; and extracting at least one visualization of the audio feature the trained neural network model relies on for classification based on the at least one saliency map.

The present disclosure relates to systems and methods for a self-adjusting corporation-wide discovery and integration feature of a machine learning system that can review a client's data store, review the labels for the various data schema, and effectively map the client's data schema to classifications used by the machine learning model. The various techniques can automatically select the features that are predictive for each individual use case (i.e., one client), effectively making a machine learning solution client-agnostic for the application developer. A weighted list of common representations of each feature for a particular machine learning solution can be generated and stored. When new data is added to the data store, a matching service can automatically detect which features should be fed into the machine-learning solution based at least in part on the weighted list. The weighted list can be updated as new data is made available to the model.

The invention discloses a cartoon-type animation generation method for a human face video image. The cartoon-type animation generation method comprises the following steps of: reading a human face video to be processed and extracting a key frame of the human face video to obtain a formula described in the specification; reading a key frame image and carrying out illustration processing to obtain an illustration image; reading the illustration image and carrying out human face feature positioning to obtain a face feature point set, and calculating the ratio of each portion of the human face according to face feature points, and carrying out feature discovery on the human face; carrying out overall deformation on the illustration image to obtain an image fti(x,y); carrying out partial exaggeration on the image fti(x,y) to obtain a final cartoon image; processing S frame by frame to obtain a key frame sequence image with a cartoon effect; reading a key frame sequence D, and carrying out middle frame generation between every two key frames by a cross decomposition method to obtain the human face video image. According to the cartoon-type animation generation method disclosed by the invention, the generated cartoon has a comedy and does not lose the original romantic charm by adopting a global-partial combination method, thus the problems of inconsistent styles and face position disproportionality of the existing cartoon are solved, and the generated illustration image is more natural and rich in artistic perception.

Disclosed systems and methods transmit and receive data encoded as optical signals that include a progression of symbols, with each symbol represented as a combination of light frequencies. Light is used as the carrier medium to limit reception to devices that are within visible range of the transmitting device. No mapping of data to symbols is required for generation of the symbol progression by the transmitter and no mapping of symbols to data for recognition of the symbols is required by the receiver. Disclosed embodiments allow arbitrary selection among a plurality of mappings between symbols sent and symbols received. Embodiments are robust to: ambient lighting conditions, differences in optical and temporal response of the first and second devices, and differences in relative orientation between the first and second devices. Embodiments enable a variety of functions including device discovery, feature discovery, beacon identification, status reporting, error readout, key exchange, and authentication.

The invention belongs to the research category of learning resource recommendation in interest recommendation research systems in social networks and relates to an automatic learning resource recommendation method based on mobile learning. The user learning interest discovery problem of a mobile user learning platform is solved by establishing models, the user learning interest problem is regardedas a track feature discovery problem of user learning recourses in time and position dimensions, users with the same resource interest are regarded as a resource user group, the next resource that users may learn is predicted in the perspectives of user learning tracks and learning characteristics of resource user groups, and the learning interests of the users can be discovered.

The subject application is directed to a document processing device, system, and method having peer device feature discovery and job routing. Electronic document data is received from an associated document processing device, which electronic document data includes data corresponding to an electronic document and instruction data corresponding to a desired document processing operation associated with the document. Data corresponding to available features of the document processing device is compared relative to received instruction data. A prompt is generated to a second document processing device based upon the output of the comparison indicating that the document processing device has insufficient capabilities for performing the desired operation. Response data is received from the second device corresponding to capabilities of the second document processing device relative to the desired operation, and the electronic document data is routed to the second device in accordance with received response data.

Automatic performance triggered model adjustment. A hierarchical feature tree is generated. Each child node's feature is more specific than its respective parent node's feature. A behavioral model comprising features of the feature tree is created and used in the operation of an advertising campaign. A degraded model feature is detected at the discovery system by comparing a performance metric of a model feature from two different time windows. The discovery system matches a node of the feature tree with the degraded feature and selects a prospective model feature from a same-level node or from a lower level node than the matching feature's node. An estimated performance metric for the prospective model feature is determined and the results are used to decide if the model should be updated to include the prospective model feature. The campaign can be operated with the automatically updated model.

The invention relates to an e-Science environment-oriented multi-domain Web text feature extracting system and method. The method comprises the following steps of: 1. making statistics on the frequentness of characters in a target text; 2. with a character as a basic processing unit, extracting character strings between the character used as a start point and a character having the frequentness of 1 and being used as a terminal point one by one; and 3. making statistics of the frequentness of each character string, and performing descending order on feature character strings according to the frequentness and outputting the feature character strings. In the invention, a non-dictionary character segmentation technology is introduced in the feature discovery of a domain text, thereby the dependence of a traditional method on a domain dictionary is effectively overcome and the portability and the practicability of the e-Science environment-oriented multi-domain Web text feature extracting system and method in multi-domain scientific data are enhanced to some extent.

The invention relates to a suspicious feature discovery method (200) based on transaction network nodes. The suspicious feature discovery method comprises the following steps: constructing a user transaction network (201) based on original transaction features; obtaining key nodes (202) from the constructed user transaction network; for each key node, performing transaction feature extraction on each transaction and forming a transaction feature set (203); performing association rule mining on the transaction feature set of each key node and aquring a transaction feature frequent set (204).

Methods, apparatus and computer readable code for determining whether a potential relay device is a relay device are provided herein. In some embodiments, first and second information elements are received from a potential relay device, which is an original source of the second information element. In order to determine whether the potential relay device is a relay device, it is determined whether a feature of an original source of the first information element and a feature of the potential relay device are features unlikely to relate to a single device, wherein a positive result of the determining is indicative that the potential relay device is a relay device. In an exemplary embodiment, a disclosed system includes an information element receiver and a feature incompatibility analyzer. Optionally, the disclosed system includes a feature discovery module, a parameter obtainer and a feature database.

A water consumption trend graphical analysis method based on a water fingerprint algorithm comprises the steps that: data are obtained through remote transmission equipment of the Internet of Things, feature discovery is carried out on the basis of the data, the data are of a waveform structure, region division is carried out on the outer side of the waveform to form 'lane', and a plurality of peak periods and low peak periods are found according to the fluctuation of the 'lane'; and meanwhile, time is added on a horizontal coordinate axis, the time is accurate to hour, data filling is carried out in a 'lane' by predicting and simulating follow-up data, and follow-up prediction data is corrected when next monitoring data is acquired. After the structure is adopted, the method has the following advantages: the water consumption trend is basically determined according to the predicted path of the lane, by adopting the graphical analysis method model, when several prediction points are exceeded to give an alarm or the sudden waveform amplitude exceeds a great prediction value, monitoring equipment judges abnormity and gives an abrupt change alarm, so that the aim of timely early warning and timely informing can be achieved.

Disclosed systems and methods transmit and receive data encoded as optical signals that include a progression of symbols, with each symbol represented as a combination of light frequencies. Light is used as the carrier medium to limit reception to devices that are within visible range of the transmitting device. No mapping of data to symbols is required for generation of the symbol progression by the transmitter and no mapping of symbols to data for recognition of the symbols is required by the receiver. Disclosed embodiments allow arbitrary selection among a plurality of mappings between symbols sent and symbols received. Embodiments are robust to: ambient lighting conditions, differences in optical and temporal response of the first and second devices, and differences in relative orientation between the first and second devices. Embodiments enable a variety of functions including device discovery, feature discovery, beacon identification, status reporting, error readout, key exchange, and authentication.

The invention relates to an e-Science environment-oriented multi-domain Web text feature extracting system and method. The method comprises the following steps of: 1. making statistics on the frequentness of characters in a target text; 2. with a character as a basic processing unit, extracting character strings between the character used as a start point and a character having the frequentness of 1 and being used as a terminal point one by one; and 3. making statistics of the frequentness of each character string, and performing descending order on feature character strings according to the frequentness and outputting the feature character strings. In the invention, a non-dictionary character segmentation technology is introduced in the feature discovery of a domain text, thereby the dependence of a traditional method on a domain dictionary is effectively overcome and the portability and the practicability of the e-Science environment-oriented multi-domain Web text feature extracting system and method in multi-domain scientific data are enhanced to some extent.

An artificially designated relationship can assist the realization of mapping, which can compress the output structure of a machine learning model. For example, an index model of a maximal entropy model can utilize machine learning embedding and mapping to generate classified output. In this way, the feature finding function of a machine learning model (for example, deep networks) and relationship, which is developed based on the people's understanding of structural properties of a problem to be solved, are cooperatively combined to compress the output structure of the model without generatingobvious precision loss. These compressed models improve the adaptability of the models in facilities or other resource limited scenes.

In an aspect of the disclosure, a method, a computer-readable medium, and an apparatus are provided. The apparatus may transmit a LMP version request PDU to a second device. The apparatus may receivea LMP version response PDU including at least one of a link layer identification, a version number, or a sub-version number associated with the second device. The apparatus may determine if one or more of the link layer identification, the version number, or the sub-version number included in the LMP version response PDU are associated with a recognized QLM. The apparatus may a first LMP encapsulated header PDU that includes a first QLMP feature request opcode to the second device when it is determined that one or more of the link layer identification, the version number, or the sub-version number included in the LMP version response PDU are associated with the recognized QLM.

Systems, methods, and computer-readable storage devices are disclosed for improved audio feature discovery using a neural network. One method including: receiving a trained neural network model, the trained neural network configured to output an audio feature classification of audio data; deconstructing the trained neural network model to generate at least one saliency map, the at least one saliency map providing a successful classification of the audio feature; and extracting at least one visualization of the audio feature the trained neural network model relies on for classification based on the at least one saliency map.

The invention provides a network public opinion topic feature extraction method and system, which firstly consider the difference between general network public opinions and large-scale network public opinions, the method is suitable for topic feature extraction of the large-scale network public opinions; secondly, a method based on a co-word network is improved, a method of combining word frequency with ITF / PDF is adopted during keyword extraction, and the weights of the keywords can be described more accurately; thirdly, the non-threshold inter-stage evolution network method adopted by the invention is simple, can retain subtle association between topics, and accords with topic evolution logic; finally, the invention provides an event-driven topic spatio-temporal feature discovery method, and address tags of the text corpora are replaced by address tags of events, so that the situation that the text corpora have no address tags can be matched.

The invention discloses a method for realizing feature discovery and identification of mobile digital communication equipment. The method comprises the following steps: the step 1, starting portable SDR equipment; the step 2, selecting a wireless signal module for loading; the step 3, capturing wireless signal data, and storing the wireless signal data into equipment for buffering., the step 4, processing and analyzing the buffered data by the analysis module, and finding the mobile digital communication equipment hardware feature code contained in the wireless signal, the step 5, identifyingwhether the hardware feature code value found in the step 4 is a specified hardware feature code value, if so, going to the step 6, and if not, going to the step 4; and the step 6, further analyzing and processing the identification result of the step 5 to collect the survival state of the mobile digital communication device and the approximate distance between the mobile digital communication device and the SDR device. According to the method, aiming at the requirements of the public security industry, the discovery and identification of various hardware feature codes in different communication protocol signals are well realized by loading software packages of corresponding protocols.

In the device state detection method and related devices provided in this application, the data processing device combines an autoencoder with multiple cyclic neural networks, and reconstructs multiple state feature sets of the device to be detected through the autoencoder, and according to The difference between multiple state feature sets and their respective reconstruction feature sets determines the health status of the device to be detected; since the features to be reconstructed of the autoencoder are encoded by multiple cyclic neural networks with different feature scales respectively Therefore, the information related to the health status of the equipment contained in the equipment status sequence can be fully explored to achieve the purpose of improving the detection accuracy.

Methods and apparatus are provided that can provide automatic feature discovery for modular active digitizer pens. A system is provided comprising a digitizer display; and a processor configured to: receive a human interface device (HID) transaction when a digitizer pen is in proximity to the digitizer display, wherein the digitizer pen comprises a pen body attached to a removable function module that supports a modular feature; extract a module identifier (ID) from the HID transaction, wherein the module ID is associated with the removable function module; and display, on the digitizer display, an informational message that concerns the modular feature, wherein the informational message is associated with the module ID. One or more applications compatible with the modular feature may be displayed, and one or more configuration tasks may be automatically executed to enable the modular feature on the system.