Abstract: Described in various embodiments herein is a technical solution directed to decomposition of time as an input for machine learning, and various related mechanisms and data structures. In particular, specific machines, computer-readable media, computer processes, and methods are described that are utilized to improve machine learning outcomes, including, improving accuracy, convergence speed (e.g., reduced epochs for training), and reduced overall computational resource requirements. A vector representation of continuous time containing a periodic function with frequency and phase-shift learnable parameters is used to decompose time into output dimensions for improved tracking of periodic behavior of a feature. The vector representation is used to modify time inputs in machine learning architectures.

Abstract: A flow of packets is communicated through a data center. The data center includes multiple racks, where each rack includes multiple network devices. A group of packets of the flow is received onto a first network device. The first device includes a neural network. The first network device generates a neural network feature vector (NNFV) based on the received packets. The first network device then sends the NNFV to a second network device. The second device uses the NNFV to determine a set of weight values. The weight values are then sent back to the first network device. The first device loads the weight values into the neural network. The neural network, as configured by the weight values, then analyzes each of a plurality of flows received onto the first device to determine whether the flow likely has a particular characteristic.

Abstract: During training of deep neural networks, a Copernican loss (LC) is designed to augment a primary loss function, for example, a standard Softmax loss, to explicitly minimize intra-class variation and simultaneously maximize inter-class variation. Copernican loss operates using the cosine distance and thereby affects angles leading to a cosine embedding, which removes the disconnect between training and testing.

Abstract: Knowledge transfer between recurrent neural networks is performed by obtaining a first output sequence from a bidirectional Recurrent Neural Network (RNN) model for an input sequence, obtaining a second output sequence from a unidirectional RNN model for the input sequence, selecting at least one first output from the first output sequence based on a similarity between the at least one first output and a second output from the second output sequence; and training the unidirectional RNN model to increase the similarity between the at least one first output and the second output.

Abstract: The invention relates to a neural network comprising: synaptic chains, each synaptic chain comprising synapses, each synapse being a spintronic resonator, the spintronic resonators being in series, each spintronic resonator having an adjustable resonance frequency, ordered layers of neurons, each neuron being a radiofrequency oscillator oscillating at its own frequency, a lower layer being connected to an upper layer by an interconnection comprising an assembly of synaptic chains connected to rectifying circuits, each resonance frequency of the assembly of synaptic chains corresponding to the frequency of a radiofrequency oscillator of the lower layer.

Abstract: Provided herein are systems and methods for an iterative approach to topic modeling and the use of web mapping technology to implement advanced spatial operators for interactive high-dimensional visualization and inference.

Abstract: An instructional design tool is provided for designing learning based applications. More specifically, the instructional design tool is configured to use captured expert knowledge for translating such knowledge into an environment used for instructional purposes. The instructional design tool includes at least one component configured to visually model a gaming scenario using recorded knowledge and graphical content defined by values associated with classes of respective models and translate the defined values into a standardized XML format.

Abstract: A computing unit for accelerating a neural network is disclosed. The computing unit may include an input unit that includes a digital-to-analog conversion unit and an analog-to-digital conversion unit that is configured to receive an analog signal from the output of a last interconnected analog crossbar circuit of a plurality of analog crossbar circuits and convert the second analog signal into a digital output vector, and a plurality of interconnected analog crossbar circuits that include the first interconnected analog crossbar circuit and the last interconnected crossbar circuits, wherein a second interconnected analog crossbar circuit of the plurality of interconnected analog crossbar circuits is configured to receive a third analog signal from another interconnected analog crossbar circuit of the plurality of interconnected crossbar circuits and perform one or more operations on the third analog signal based on the matrix weights stored by the crosspoints of the second interconnected analog crossbar.

Abstract: A building management system (BMS) includes sensors that measure time series values of building variables and a deterministic model generator that uses historical values for the time series of building variables to train a deterministic model that predicts deterministic values for the time series. The BMS includes a stochastic model generator that uses differences between actual values for the time series and the predicted deterministic values to train a stochastic model that predicts a stochastic value for the time series. The BMS includes a forecast adjuster that adjusts the predicted deterministic values using the predicted stochastic value to generate an adjusted forecast for the time series. The BMS includes a demand response optimizer that uses the adjusted forecast to generate an optimal set of control actions for building equipment of the BMS. The building equipment operate to affect the building variables.

Abstract: Methods, systems, and apparatus, including computer programs encoded on computer storage media, for training machine learning systems. One of the methods includes receiving a plurality of training examples; and training a machine learning system on each of the plurality of training examples to determine trained values for weights of a machine learning model, wherein training the machine learning system comprises: assigning an initial value for a regularization penalty for a particular weight for a particular feature; and adjusting the initial value for the regularization penalty for the particular weight for the particular feature during the training of the machine learning system.

Abstract: A system and method provides video recommendations for a target video in a video sharing environment. The system selects one or more videos that are on one or more video playlists together with the target video. The video co-occurrence data of the target video associates the target video and another video on one or more same video playlists and frequency of the target video and another video on the video playlists is computed. Based on the video co-occurrence data of the target video, one or more co-occurrence videos are selected and ranked based on the video co-occurrence data of the target video. The system selects one or more videos from the co-occurrence videos as video recommendations for the target video.

Abstract: Methods, systems, and apparatus, including computer programs encoded on computer storage media, for predicting the accuracy of user submissions. One of the methods includes receiving, from a user, an update to an attribute of an entity related to a topic. If the user is determined to be reliable relative to the topic based on user profile data of the user, the knowledge base is updated with the update to the attribute of the entity.

Abstract: A machine learning device, which learns an operation program of a robot, includes a state observation unit which observes as a state variable at least one of a shaking of an arm of the robot and a length of an operation trajectory of the arm of the robot; a determination data obtaining unit which obtains as determination data a cycle time in which the robot performs processing; and a learning unit which learns the operation program of the robot based on an output of the state observation unit and an output of the determination data obtaining unit.

Abstract: A building management system (BMS) includes sensors that measure time series values of building variables and a deterministic model generator that uses historical values for the time series of building variables to train a deterministic model that predicts deterministic values for the time series. The BMS includes a stochastic model generator that uses differences between actual values for the time series and the predicted deterministic values to train a stochastic model that predicts a stochastic value for the time series. The BMS includes a forecast adjuster that adjusts the predicted deterministic values using the predicted stochastic value to generate an adjusted forecast for the time series. The BMS includes a demand response optimizer that uses the adjusted forecast to generate an optimal set of control actions for building equipment of the BMS. The building equipment operate to affect the building variables.

Abstract: A method and system for controlling access to an Internet resource is disclosed herein. When a request for an Internet resource, such as a Web site, is transmitted by an end-user of a LAN, a security appliance for the LAN analyzes a reputation index for the Internet resource before transmitting the request over the Internet. The reputation index is based on a reputation vector which includes a plurality of factors for the Internet resource such as country of domain registration, country of service hosting, country of an internet protocol address block, age of a domain registration, popularity rank, internet protocol address, number of hosts, to-level domain, a plurality of run-time behaviors, JavaScript block count, picture count, immediate redirect and response latency. If the reputation index for the Internet resource is at or above a threshold value established for the LAN, then access to the Internet resource is permitted.

Abstract: A multi-factor probabilistic model evaluates user input to determine if the user input was intended for an on-screen user interface control. When user input is received, a probability is computed that the user input was intended for each on-screen user interface control. The user input is then associated with the user interface control that has the highest computed probability. The probability that user input was intended for each user interface control may be computed utilizing a multitude of factors including the probability that the user input is near each user interface control, the probability that the motion of the user input is consistent with the user interface control, the probability that the shape of the user input is consistent with the user interface control, and that the size of the user input is consistent with the user interface control.

Abstract: A classifier training method includes detecting error data from training data; and training a classifier configured to detect an object based on the error data.

Abstract: An autonomous thinking pattern generator including a pattern converter configured to convert input information to patterns, the input information including image information, sound information or language, a pattern recorder configured to record the patterns, a pattern controller configured to set and change the patterns, and form connective relations between the patterns, and an information analyzer configured to evaluate values of the input information is provided. The pattern recorder is configured to record the patterns corresponding to the input information which is determined as worthy by the information analyzer autonomously.

Abstract: Methods, systems, and apparatus, including computer programs encoded on computer storage media, for routing events of an event stream in a stream processing system. One of the methods includes receiving, by a router, an event stream of events; identifying, for each event, by the router, a respective partition of context data that includes context data related to the event and providing the event to a respective local modeler that stores the partition of context data identified for the event in operational memory of the local modeler; processing, by each local modeler, events received from the router and aggregating information associated with each event to generate aggregated information; providing, by one or more of the local modelers, to a central modeler, the respective aggregated information; and determining, by the central modeler, a plurality of parameters of a machine learning model using the received aggregated information.

Abstract: A classifier is computed as follows. For a first set of values of primary field(s) of primary data instances of a labeled primary training dataset, a second set(s) of secondary fields of unclassified second data instances of secondary dataset(s) is identified. First set of values are matched to corresponding values in respective secondary field(s), and linked to other secondary fields of respective secondary data instance(s) of the respective matched secondary field. A set of classification features is generated. Each including: (i) condition(s), and (ii) a value selected from the linked other secondary fields of the respective secondary data instance(s) of the respective matched secondary field(s). The respective classification feature outputs a binary value computed by the condition(s) that compares between the value selected from the other linked secondary fields and a new received data instance. A classifier is computed according to a selected subset of the classification features.