Abstract: A neural device to which a conditioned response function is imparted and a driving method thereof are disclosed. Quantum dots and a polymer insulating layer are formed between upper and lower electrodes. Conductive filaments are formed at interfaces between the quantum dots and the polymer insulating layer. When a positive pulse, which is an unconditioned stimulus signal, is applied, the conductive filaments are formed, and a low resistance state is implemented. As the number of applications of a negative pulse, which is a conditioned stimulus signal, increases, the neural device is switched from a high resistance state to the low resistance state. Through this, the neural device having learning ability for the conditioned stimulus signal may be implemented and driven.

Abstract: Aspects of reusing neural network instructions are described herein. The aspects may include a computing device configured to calculate a hash value of a neural network layer based on the layer information thereof. A determination unit may be configured to determine whether the hash value exists in a hash table. If the hash value is included in the hash table, one or more neural network instructions that correspond to the hash value may be reused.

Abstract: Aspects of a neural network operation device are described herein. The aspects may include a matrix element storage module configured to receive a first matrix that includes one or more first values, each of the first values being represented in a sequence that includes one or more bits. The matrix element storage module may be further configured to respectively store the one or more bits in one or more storage spaces in accordance with positions of the bits in the sequence. The aspects may further include a numeric operation module configured to calculate an intermediate result for each storage space based on one or more second values in a second matrix and an accumulation module configured to sum the intermediate results to generate an output value.

Abstract: A hardware acceleration component is provided for implementing a convolutional neural network. The hardware acceleration component includes an array of N rows and M columns of functional units, an array of N input data buffers configured to store input data, and an array of M weights data buffers configured to store weights data. Each of the N input data buffers is coupled to a corresponding one of the N rows of functional units. Each of the M weights data buffers is coupled to a corresponding one of the M columns of functional units. Each functional unit in a row is configured to receive a same set of input data. Each functional unit in a column is configured to receive a same set of weights data from the weights data buffer coupled to the row. Each of the functional units is configured to perform a convolution of the received input data and the received weights data, and the M columns of functional units are configured to provide M planes of output data.

Abstract: Technologies for performing in-memory training data augmentation for artificial intelligence include a memory comprising media access circuitry connected to a memory media. The media access circuitry is to obtain an input training data set that includes an initial amount of data samples that are usable to train a neural network. The media access circuitry is further to produce, from the input training data set, an augmented training data set with more data samples than the input training data set.

Abstract: Inspired by the processing methods of biologic brains, we construct a network of multiple configurable non-volatile memory arrays connected with bus-lines as a neuromorphic code processor for code processing. In contrast to the Von-Neumann computing architectures applying the multiple computations for code vector manipulations, the neuromorphic code processor of the invention processes codes according to their configured codes stored in the non-volatile memory arrays. Similar to the brain processor, the neuromorphic code processor applies the one-step feed-forward processing in parallel resulting in a dramatic power reduction compared with the computational methods in the conventional computer processors.

Abstract: A hybrid neuromorphic computing device is provided, in which artificial neurons include light-emitting devices that provide weighted sums of inputs as light output. The output is detected by a photodetector and converted to an electrical output. Each neuron may receive output from one or more other neurons as initial input, allowing for high degrees of fan-out and fan-in, including true broadcast-to-all functionality.

Abstract: Systems and methods for training neural networks. One embodiment is a system that includes a memory configured to store samples of training data for a Deep Neural Network (DNN), and a distributor. The distributor identifies a plurality of work servers provisioned for training the DNN by processing the samples via a model of the DNN, receives information indicating Graphics Processing Unit (GPU) processing powers at the work servers, determines differences in the GPU processing powers between the work servers based on the information, and allocates the samples among the work servers based on the differences.

Abstract: Strategies for improved neural network fine tuning. Parameters of the task-specific layer of a neural network are initialized using approximate solutions derived by a variant of a linear discriminant analysis algorithm. One method includes: inputting training data into a deep neural network having an output layer from which output is generated in a manner consistent with one or more classification tasks; evaluating a distribution of the data in a feature space between a hidden layer and the output layer; and initializing, non-randomly, the parameters of the output layer based on the evaluated distribution of the data in the feature space.

Abstract: According to the present disclosure, an apparatus includes at least one processor; and at least one memory including computer program code. The at least one memory and the computer program code are configured, with the at least one processor, to cause the apparatus to receive media content for streaming to a user device; to train a neural network to be overfitted to at least a first portion of the media content; and to send the trained neural network and the first portion of the media content to the user equipment. In addition, another apparatus includes at least one processor; and at least one memory including computer program code. The at least one memory and the computer program code are configured, with the at least one processor, to cause the apparatus to receive at least a first portion of media content and a neural network trained to be overfitted to the first portion of the media content; and to process the first portion of the media content using the overfitted neural network.

Abstract: Behavioral verification of user identity includes building a deep neural network for keystroke-based behavioral verification of user identity. The building includes receiving recorded keystroke events, each such recorded keystroke event including (i) an indication of whether the recorded keystroke event is a key press or a key release, (ii) a key identifier of the respective key pressed or released, and (iii) a timestamp of the recorded keystroke event. The building further includes performing pre-processing of the recorded keystroke events to provide data structures representing sequential key events for processing by a deep neural network to extract local patterns, and training the deep neural network using the data structures. The method also includes providing the trained deep neural network for keystroke-based behavioral verification of user identity based on determinate vectors output from the trained deep neural network.

Abstract: Behavioral verification of user identity includes building a deep neural network for gait-based behavioral verification of user identity. The building includes receiving movement data describing movement, in multiple dimensions, of computer system(s) of user(s), the movement data including sensor data acquired from sensor(s) of the computer system(s). The building further includes performing pre-processing of the movement data to provide processed movement data for processing by a deep neural network to extract local patterns, and training the deep neural network using the processed movement data. The method also includes providing the trained deep neural network for gait-based behavioral verification of user identity based on determinate vectors output from the trained deep neural network.

Abstract: A learning use data set showing relationships among an engine speed, an engine load rate, an air-fuel ratio of the engine, an ignition timing of the engine, an HC or CO concentration of exhaust gas flowing into an exhaust purification catalyst and a temperature of the exhaust purification catalyst is acquired. The acquired engine speed, engine load rate, air-fuel ratio of the engine, ignition timing of the engine, and HC or CO concentration of the exhaust gas flowing into the exhaust purification catalyst are used as input parameters of a neural network and the acquired temperature of the exhaust purification catalyst is used as training data to learn a weight of the neural network. The learned neural network is used to estimate the temperature of the exhaust purification catalyst.

Abstract: Memristive learning concepts for neuromorphic circuits are described. In one example case, a neuromorphic circuit includes a first oscillatory-based neuron that generates a first oscillatory signal, a diode that rectifies the first oscillatory signal, and a synapse coupled to the diode and including a long-term potentiation (LTP) memristor arranged in parallel with a long-term depression (LTD) memristor. The circuit further includes a difference amplifier coupled to the synapse that generates a difference signal based on a difference between output signals from the LTP and LTD memristors, and a second oscillatory-based neuron electrically coupled to the difference amplifier that generates a second oscillatory signal based on the difference signal. The circuit also includes a feedback circuit that provides a feedback signal to the LTP and LTD memristors based on a difference or error between a target signal and the second oscillatory signal.

Abstract: Aspects of activation function computation for neural networks are described herein. The aspects may include a search module configured to receive an input value. The search module may be further configured to identify a data range based on the received input value and an index associated with the data range. Meanwhile, a count value may be set to one. Further, the search module may be configured to identify a slope value and an intercept value that correspond to the input value. A computation module included in the aspects may be configured to calculate an output value based on the slope value, the intercept value and the input value. In at least some examples, the process may be repeated to increase the accuracy of the result until the count of the repetition reaches the identified index.

Abstract: This method comprises: obtaining a first module for labelling images by machine learning on the basis of a first training corpus; obtaining a second training corpus from the first training corpus, by replacing, in the first training corpus, each of a portion of first labels by a replacement label, two first labels being replaced by one and the same replacement label; obtaining a second module for labelling images by machine learning on the basis of the second training corpus; obtaining the system for labelling images comprising: a first upstream module obtained from a portion of the first module, a second upstream module obtained from a portion of the second module and a downstream module designed to provide a labelling of an image on the basis of first descriptive data provided by the first upstream module and of second descriptive data provided by the second upstream module.

Abstract: Aspects of data modification for neural networks are described herein. The aspects may include a data modifier configured to receive input data and weight values of a neural network. The data modifier may include an input data configured to modify the received input data and a weight modifier configured to modify the received weight values. The aspects may further include a computing unit configured to calculate one or more groups of output data based on the modified input data and the modifier weight values.

Abstract: The system described herein can include neural networks with noise-injection layers. The noise-injection layers can enable the neural networks to be trained such that the neural networks are able to maintain their classification and prediction performance in the presence of noisy data signals. Once trained, the parameters from the neural networks with noise-injection layers can be used in the neural networks of systems that include resistive random-access memory (ReRAM), memristors, or phase change memory (PCM), which use analog signals that can introduce noise into the system. The use of ReRAM, memristors, or PCM can enable large-scale parallelism that improves the speed and computational efficiency of neural network training and classification. Using the parameters from the neural networks trained with noise-injection layers, enables the neural networks to make robust predictions and calculations in the presence of noisy data.

Abstract: A promotion value model uses deep neural networks to learn to calculate the promotion value of a commercial brand. The model determines and reports the promotion value of a plurality of electronic media files each containing at least one commercial brand indicator. The learned model identifies the electronic media files and determines at least one context for each of the at least one commercial brand indicators. Promotion value is modeled with a deep neural network that maps the context for each of the commercial brand indicators to feature vectors mapped to an input layer of the neural network. Network parameters are learned to indicate relative weighted values between transitions of the layers of the neural network.

Abstract: To efficiently process a programming problem including a function defined piecewise without having the differentiability and continuity of the function expressing the problem or spatial continuity as prerequisites, a non-linear programming problem processing device is provided with: a non-linear programming problem input unit; a provisional solution generation unit that produces a solution obtained in a certain region of the non-linear programming problem as a provisional solution; a solution candidate generation unit that produces a solution obtained in a nearby region of the provisional solution as a solution candidate; a provisional solution update unit that updates the solution candidate in accordance with the result of comparison of the provisional solution and the solution candidate; an end determination unit that determines the end of the process using a provisional solution improvement degree and/or the number of times of generation of the solution candidate; and a non-linear programming problem solut

Abstract: Concepts and technologies disclosed herein are directed to the optimization of over-the-air (“OTA”) file distribution for connected cars based upon a heuristic scheduling algorithm. A schedule provided by the heuristic scheduling algorithm is designed to distribute OTA data flow to connected cars over the network (geographically) and over a scheduling time horizon (timely), and is capable of reducing the negative impact of OTA file updates on overall wireless network performance. This schedule is created based upon historical statistics associated with connected car driving patterns and simulations of connected car-specific OTA traffic over the network. By leveraging connected cars that connect to different cells at different times based upon driving patterns, the heuristic scheduling algorithm is effective in reducing OTA impact on the network.

Abstract: Examples of analyzing documents are defined. In an example, a request to analyze a document may be received. A knowledge model corresponding to a guideline associated with the document may be obtained. The knowledge model may include at least one of a hypothetical question and a logical flow to determine an inference to the hypothetical question. The hypothetical question relates to an element of the guideline. Based on the knowledge model, data from the document may be extracted for analysis using an artificial intelligence (AI) component. The Ai component may be configured to extract and analyze data, based on the knowledge model. Based on the analysis, a report indicating whether the document falls within a purview of the guideline may be generated.

Abstract: A behavior prediction device comprising: a moving object behavior detection unit configured to detect moving object behavior, a behavior prediction model database that stores a behavior prediction model, a behavior prediction calculation unit configured to calculate a behavior prediction of the moving object using the behavior prediction model, a prediction deviation determination unit configured to determine whether a prediction deviation occurs based on the behavior prediction and a detection result of the moving object behavior corresponding to the behavior prediction, a deviation occurrence reason estimation unit configured to estimate a deviation occurrence reason when determination is made that the prediction deviation occurs, and an update necessity determination unit configured to determine a necessity of an update of the behavior prediction model database based on the deviation occurrence reason when the determination is made that the prediction deviation occurs.

Abstract: Disclosed is a computer-implemented method for generating a prediction model. The model can be for use in processing machine event data to predict behavior of a plurality of industrial machines under supervision. The prediction model can be configured to determine current and future states of the industrial machines. The method can include: extracting event features from event codes and structuring the event features into feature vectors; and generating the prediction model by clustering the feature vectors into a plurality of vector clusters, the vector clusters being assigned to respective machine states. The prediction model can be constructed based on event data from a first industrial machine and be applied to control an operating state of a second industrial machine.

Abstract: An approach is provided for performing an item tracking operation. The item tracking operation includes defining a knowledge model where the knowledge model correlates usage of an item away from a parked location with other locations visited by a user. The item tracking operation also includes tracking the item when the item is removed from the parked location where the tracking includes determining when the item is moved to a particular location. The item tracking operation also includes determining whether the item is returned to the parked location and notifying the user when the item was not returned to the parked location where the notifying calculates a probability that the item was left behind at the particular location.

Abstract: Embodiments of the present disclosure disclose a method and apparatus for recommending entity. A method for recommending entity includes: acquiring a candidate entity set associated with a to-be-searched entity, in response to receiving a user's search request for an entity; inputting the candidate entity set into a pre-trained ranking model to obtain a candidate entity sequence; and selecting a candidate entity from the candidate entity sequence and recommending the selected candidate entity to the user. The ranking model ranks the candidate entity set based on at least one of: a degree of correlation between each candidate entity in the candidate entity set and the to-be-searched entity; a degree of interest of the user in the each candidate entity in the candidate entity set; and a degree of expectation of the user for the each candidate entity in the candidate entity set.

Abstract: A cognitive learning method comprising: receiving data from a plurality of data sources; processing the data from the plurality of data sources to perform a cognitive learning operation, the processing being performed via a cognitive inference and learning system, the cognitive learning operation comprising a plurality of cognitive learning operation lifecycle phases, the cognitive learning operation applying a cognitive learning technique to generate a cognitive learning result; and, updating a destination based upon the cognitive learning result.

Abstract: Embodiments of the invention are directed to systems, methods, and computer program products for dynamic conditioning for advanced misappropriation protection. The system identifies new/emerging misappropriations and profiles them into synthetic data streams via distribution of the misappropriation into a separate channel and allowing processing. Processing the misappropriation allows for analytical data generation and synthetic misappropriation generation. The synthetic stream is injected into a matrix of learning engines for learning of the misappropriation. The learning engines monitor and report each other and are arranged in an architecture form for an implicit internal feedback loop and an explicit external feedback loop.

Abstract: An application performance analyzer adapted to analyze the performance of one or more applications running on IT infrastructure, comprises: a data collection engine collecting performance metrics for one or more applications running on the IT infrastructure; an anomaly detection engine analyzing the performance metrics and detecting anomalies, i.e. performance metrics whose values deviate from historic values with a deviation that exceeds a predefined threshold; a correlation engine detecting dependencies between plural anomalies, and generating anomaly clusters, each anomaly cluster consisting of anomalies that are correlated through one or more of the dependencies; a ranking engine ranking anomalies within an anomaly cluster; and a source detection engine pinpointing a problem source from the lowest ranked anomaly in an anomaly cluster.

Abstract: In some embodiments, a computing system computes, with a state prediction model, probabilities of transitioning from a click state represented by interaction data to various predicted next states. The computing system computes an interface experience metric for the click with an experience valuation model. To do so, the computing system identifies base values for the click state and the predicted next states. The computing system computes value differentials for between the click state's base value and each predicted next state's base value. Value differentials indicate qualities of interface experience. The computing system determines the interface experience metric from a summation that includes the current click state's base value and the value differentials weighted with the predicted next states' probabilities.

Abstract: A storm damage response system includes a storm ensemble database that stores ensemble forecast models associated with respective potential storm paths of a given storm across a geographic area and an inventory database that stores inventory data associated with location and characteristics of power-providing equipment and characteristics of a consumer population in the geographic area. A storm damage model algorithm generates a storm response plan comprising an operational procedure for repairing or maintaining power transmission and distribution electric systems to mitigate storm damage impact based on generating a probabilistic model for each of the ensemble forecast models based on the inventory data and calculating a statistical impact value associated with the probabilistic model based on an aggregate of iterative probabilistic simulations for the respective ensemble forecast model.

Abstract: A relational event history is determined based on a data set, the relational event history including a set of relational events that occurred in time among a set of actors. Data is populated in a probability model based on the relational event history, where the probability model is formulated as a series of conditional probabilities that correspond to a set of sequential decisions by an actor for each relational event, where the probability model includes one or more statistical parameters and corresponding statistics. A baseline communications behavior for the relational event history is determined based on the populated probability model, and departures within the relational event history from the baseline communications behavior are determined.

Abstract: A relational event history is determined based on a data set, the relational event history including a set of relational events that occurred in time among a set of actors. Data is populated in a probability model based on the relational event history, where the probability model is formulated as a series of conditional probabilities that correspond to a set of sequential decisions by an actor for each relational event, where the probability model includes one or more statistical parameters and corresponding statistics. A baseline communications behavior for the relational event history is determined based on the populated probability model, and departures within the relational event history from the baseline communications behavior are determined.

Abstract: A system and method to determine building thermal performance parameters through empirical testing is described. The parameters can be formulaically applied to determine fuel consumption and indoor temperatures. To generalize the approach, the term used to represent furnace rating is replaced with HVAC system rating. As total heat change is based on the building's thermal mass, heat change is relabeled as thermal mass gain (or loss). This change creates a heat balance equation that is composed of heat gain (loss) from six sources, three of which contribute to heat gain only. No modifications are required for apply the empirical tests to summer since an attic's thermal conductivity cancels out and the attic's effective window area is directly combined with the existing effective window area. Since these tests are empirically based, the tests already account for the additional heat gain associated with the elevated attic temperature and other surface temperatures.

Abstract: Among other things, an apparatus comprises quantum units; and couplers among the quantum units. Each coupler is configured to couple a pair of quantum units according to a quantum Hamiltonian characterization of the quantum by the coupler.

Abstract: A system and method for batched, supervised, in-situ machine learning classifier retraining for malware identification and model heterogeneity.

Abstract: The embodiments disclosed in this document are directed to an AI-enabled microgrid and DER planning platform that uses AI methods and takes into account cost calculations, emission calculations, technology investments and operation. In an embodiment, the computing platform is deployed on a network (cloud computing platform) that can be accessed by a variety of stakeholders (e.g., investors, technology vendors, energy providers, regulatory authorities). In an embodiment, the planning platform implements machine learning (e.g., neural networks) to estimate various planning parameters, where the neural networks are trained on observed data from real-world microgrid/minigrid and DER projects.

Abstract: Balancing content distribution between a machine learning model and a statistical model provides a baseline assurance in combination with the benefits of a well-trained machine learning model for content selection. In some implementations, a server receiving requests for a content item assigns a first proportion of the received requests to a first group and assigns remaining requests to a second group. The server uses a machine learning model to select variations of the requested content item for responding to requests assigned to the first group and uses a statistical model to select content variations for requests assigned to the second group. The server obtains performance information, e.g., acceptance rates for the different variations, and compares performance of the different models used for content selection. Audience share assigned to the machine learning model is increased when it outperforms the statistical model and decreased when it underperforms the statistical model.

Abstract: A method for machine learning performed by a computer includes: (i) executing a first process that includes executing machine learning on weight values corresponding to multiple functions to be used to calculate similarities between items forming pairs and included in first and second data included in a teacher data item for each of the pairs of items based on the teacher data item stored in a memory; and (ii) executing a second process that includes identifying evaluation functions to be used to calculate the similarities between the items forming the pairs based on the multiple functions and the weight values corresponding to the multiple functions.

Abstract: Among other things, motion data is acquired from a device in a vehicle during a trip. The motion data is applied to a trained classifier to produce a commercial classification of the vehicle.

Abstract: One or more machine-learning models are trained and employed to predict test coverage and test data volume. Input features for the one or more machine-learning models comprise the test configuration features and the design complexity features. The training data are prepared by performing test pattern generation and circuit design analysis. The design complexity features may comprise testability, X-profiling, clock domains, power domains, design-rule-checking warnings, or any combination thereof.

Abstract: Provided is an artificial intelligence system. The system includes a first sensor configured to generate a first sensing signal during a sensing time, a second sensor disposed adjacent to the first sensor and configured to generate a second sensing signal during the sensing time, a pre-processing unit configured to select valid data according to a magnitude of a differential signal generated based on a difference between the first sensing signal and the second sensing signal, and an artificial intelligence module configured to analyze the valid data to generate result data.

Abstract: Methods and systems for automated tuning of a service configuration are disclosed. An optimal configuration for a test computer is selected by performing one or more load tests using the test computer for each of a plurality of test configurations. The performance of a plurality of additional test computers configured with the optimal configuration is automatically determined by performing additional load tests using the additional test computers. A plurality of production computers are automatically configured with the optimal configuration if the performance of the additional test computers is improved with the optimal configuration.

Abstract: A method, system and computer product for performing storage maintenance is described. A training set for storage volume reclamation is received. The training set for storage volume reclamation contains sets of storage parameters for storage volumes and corresponding user decisions whether the storage volumes are reclaimable. The training set is used to train a machine learning system to recognize reclaimable candidate storage volumes. The trained machine learning system is used to determine that a candidate storage volume for reclamation is likely a reclaimable storage volume.

Abstract: In some embodiments, the present invention provides for a computer system which includes a content database storing initial content data and a vocabulary data set; a processor configured to applying a machine learning model to transform the initial content data into a N-dimensional vector space; self-training the machine learning model based on the vocabulary data set; applying a clustering technique to the N-dimensional vector space to generate a cluster model of clusters, where each cluster includes a plurality of word representations; associating each cluster with a cluster identifier; obtaining subsequent content data; associating each data element of the subsequent content data with each cluster to generate a content data cluster mapping model; continuously tracking, for each user, each respective cluster identifier of each respective cluster associated with each action performed by each user with each data element to continuously self-adapt each user-specific, time-specific dynamic cluster mapping model

Abstract: Embodiments are directed towards automatically learning user behavioral patterns when interacting with messages and based on the learned patterns, suggesting one or more predicted actions that a user might take in response to receiving subsequent message. One or more classifiers are trained and employed to predict one or more actions that a user might take in response to receiving the message. In one embodiment, the one or more predicted actions are provided suggested to the user as an action the user might take on the message. Messages may be rank ordered within a given suggested action based on a confidence level of the prediction.

Abstract: A method of qubits, a room temperature quantum computing and a system including a controller, a readout, a resistor and a storage are disclosed. The shape and area of each qubits and the pattern of qubit array may be defined by a pattern on a mask simultaneously to control correlations among qubits. The configuration of qubit correlation may be designed three-dimensionally by stacking layers including arrays of qubits. The external generator may be included in another layer stacked with the layers including the arrays of qubits. The qubit may comprise a band structure having a spin-less ground state and a first excited state with spin. The first excited state may not be split for a retention time even under the external field which can influence a spin. The configuration of qubit correlations may be tuned by considering this retention time and an error correction code in quantum computation.

Abstract: Systems and devices for the evaluation and analysis of computing system data for anomaly detection and processing are disclosed. In an example, operations to detect anomalies include: obtaining a source event stream of data produced from operation of a computing system; converting the source event stream into a frequency signal; identifying an estimated seasonality from the frequency signal; scaling the estimated seasonality to a target event stream produced from operation of the computing system; identifying anomalies of a principal vector of the target event stream, based on deviation from the estimated seasonality; and causing the computing system to perform an action based on the identified anomalies. In a further example, the operations include scaling the estimated seasonality to a subset of the target event stream indicating minor vectors; and identifying anomalies of the minor vectors, based on deviation of the subset of data from the estimated seasonality.

Abstract: Systems and methods are provided for training a machine learned model on a large number of devices, each device acquiring a local set of training data without sharing data sets across devices. The devices train the model on the respective device's set of training data. The devices communicate a parameter vector from the trained model asynchronously with a parameter server. The parameter server updates a master parameter vector and transmits the master parameter vector to the respective device.

Abstract: To train models, training data is needed. As personal data changes over time, the training data can get stale, obviating its usefulness in training the model. Embodiments deal with this by developing a database with a running log specifying how each person's data changes at the time. When data is ingested, it may not he normalized. To deal with this, embodiments clean the data to ensure the ingested data fields are normalized. Finally, the various tasks needed to train the model and solve for accuracy of personal data can quickly become cumbersome to a computing device. They can conflict with one another and compete inefficiently for computing resources, such as processor power and memory capacity. To deal with these issues, a scheduler is employed to queue the various tasks involved.