Abstract: An audio visual haptic signal reconstruction method includes first utilizing a large-scale audio-visual database stored in a central cloud to learn knowledge, and transferring same to an edge node; then combining, by means of the edge node, a received audio-visual signal with knowledge in the central cloud, and fully mining semantic correlation and consistency between modals; and finally fusing the semantic features of the obtained audio and video signals and inputting the semantic features to a haptic generation network, thereby realizing the reconstruction of the haptic signal. The method effectively solves the problems that the number of audio and video signals of a multi-modal dataset is insufficient, and semantic tags cannot be added to all the audio-visual signals in a training dataset by means of manual annotation. Also, the semantic association between heterogeneous data of different modals are better mined, and the heterogeneity gap between modals are eliminated.

Abstract: Systems and methods for flexible parameter sharing for multi-task learning are provided. A training method can include obtaining a test input, selecting a particular task from one or more tasks, and training a multi-task machine-learned model for the particular task by performing a forward pass using the test input and one or more connection probability matrices to generate a sample distribution of test outputs, training the components of the machine-learned model based at least in part on the sample distribution, and performing a backwards pass to train a connection probability matrix of the multi-task machine-learned model using a straight-through Gumbel-softmax approximation.

Abstract: A computer-implemented method for automatically determining a neural network architecture represents a neural network architecture as a data structure defining a hierarchical set of directed acyclic graphs in multiple levels. Each graph has an input, an output, and a plurality of nodes between the input and the output. At each level, a corresponding set of the nodes are connected pairwise by directed edges which indicate operations performed on outputs of one node to generate an input to another node. Each level is associated with a corresponding set of operations. At a lowest level, the operations associated with each edge are selected from a set of primitive operations. The method includes repeatedly generating new sample neural network architectures, and evaluating their fitness. The modification is performed by selecting a level, selecting two nodes at that level, and modifying, removing or adding an edge between those nodes according to operations associated with lower levels of the hierarchy.

Abstract: In one implementation, a method is implemented by a neural network device and includes inputting a representation of topological structures in patterns of activity in a source neural network, wherein the activity is responsive to an input into the source neural network, processing the representation, and outputting a result of the processing of the representation. The processing is consistent with a training of the neural network to process different such representations of topological structures in patterns of activity in the source neural network.

Abstract: A system assigns experts of a mixture-of-experts artificial intelligence model to processing devices in an automated manner. The system includes an orchestrator component that maintains priority data that stores, for each of a set of experts, and for each of a set of execution parameters, ranking information that ranks different processing devices for the particular execution parameter. In one example, for the execution parameter of execution speed, and for a first expert, the priority data indicates that a central processing unit (“CPU”) executes the first expert faster than a graphics processing unit (“GPU”). In this example, for the execution parameter of power consumption, and for the first expert, the priority data indicates that a GPU uses less power than a CPU. The priority data stores such information for one or more processing devices, one or more experts, and one or more execution characteristics.

Abstract: A computing device selects new test configurations for testing software. (A) First test configurations are generated using a random seed value. (B) Software under test is executed with the first test configurations to generate a test result for each. (C) Second test configurations are generated from the first test configurations and the test results generated for each. (D) The software under test is executed with the second test configurations to generate the test result for each. (E) When a restart is triggered based on a distance metric value computed between the second test configurations, a next random seed value is selected as the random seed value and (A) through (E) are repeated. (F) When the restart is not triggered, (C) through (F) are repeated until a stop criterion is satisfied. (G) When the stop criterion is satisfied, the test result is output for each test configuration.

Abstract: A model building device and a loading disaggregation system are provided. The model building device disaggregates total aggregated data outputted from a total electricity meter and measured during a unit processing period. The model building device includes a usage pattern-analyzing module, an information mapping module, and a time series-analyzing module. After receiving the total aggregated data, the usage pattern-analyzing module analyzes the total aggregated data based on detection conditions and generates usage pattern information accordingly. The information mapping module maps the usage pattern information to form encoded data. The time series-analyzing module analyzes time correlation of the encoded data to generate synthesized simulation data.

Abstract: A computing device selects new test configurations for testing software. Software under test is executed with first test configurations to generate a test result for each test configuration. Each test configuration includes a value for each test parameter where each test parameter is an input to the software under test. A predictive model is trained using each test configuration of the first test configurations in association with the test result generated for each test configuration based on an objective function value. The predictive model is executed with second test configurations to predict the test result for each test configuration of the second test configurations. Test configurations are selected from the second test configurations based on the predicted test results to define third test configurations. The software under test is executed with the defined third test configurations to generate the test result for each test configuration of the third test configurations.

Abstract: A soldering process method includes steps of: establishing a material component database; establishing a working parameter database; analyzing material and component characteristics required for a new soldering process; comparing the characteristics with information in the material component database; selecting operating parameters corresponding to the material and component characteristics similar to those required for the new soldering process; performing the soldering process using the operating parameters corresponding to the material and component characteristics similar to those required for the new soldering process; measuring and recording the soldering process execution information and the final product information; determining whether the final product of the solder process meets the quality control requirements; using the machine learning method to fit the soldering process execution information and the final product information of the solder process to get the operating parameters for the next sol

Abstract: A transition control unit detects, when stochastically determining based on a temperature, energy changes, and a random number whether to accept any of a plurality of state transitions according to a relative relationship between the energy changes and thermal excitation energy, a minimum value among the energy changes. The transition control unit then subtracts, when the minimum value is positive, an offset obtained by multiplying the minimum value by a value M that is greater than 0 and less than or equal to 1 from each of the energy changes corresponding to the plurality of state transitions.

Abstract: A hardware neural network system includes an input buffer for input neurons (Nbin), an output buffer for output neurons (Nbout), and a third buffer for synaptic weights (SB) connected to a Neural Functional Unit (NFU) and a control logic (CP) for performing synapses and neurons computations. The NFU pipelines a computation into stages, the stages including weight blocks (WB), an adder tree, and a non-linearity function.

Abstract: An automated system for recommending predictor variable values for improving predictive outcomes of a predictive model is provided. The automated system recommends appropriate predictor variable values for changeable predictor variables that improve a predictive outcome of the predictive model by (i) computing predictive outcomes for each input record during a batch ETL process and (ii) determining appropriate predictor variable values that lead to improved predictive outcomes, using the code generated extended ETL jobs updated to perform rescoring using a combination of different values of the changeable predictor variables while honoring constraints between the changeable predictor variables, or by enabling an end user to perform said rescoring by changing values of the changeable predictor variables on the fly to determine most suitable predictor variable values that lead to improved predictive outcomes.

Abstract: Techniques are disclosed for the generation of adversarial training data through sequence perturbation, for a deep learning network to perform event sequence analysis. A methodology implementing the techniques according to an embodiment includes applying a long short-term memory attention model to an input data sequence to generate discriminative sequence periods and attention weights associated with the discriminative sequence periods. The attention weights are generated to indicate the relative importance of data in those discriminative sequence periods. The method further includes generating perturbed data sequences based on the discriminative sequence periods and the attention weights. The generation of the perturbed data sequences employs selective filtering or conservative adversarial training, to preserve perceptual similarity between the input data sequence and the perturbed data sequences.

Abstract: Systems and methods for the annotation of source data in accordance with embodiments of the invention are disclosed. In one embodiment, a data annotation server system obtains a set of source data, provides at least one subset of source data to at least one annotator device, obtains a set of annotation data from the at least one annotator device for each subset of source data, classifies the source data based on the annotation data using a machine classifier for each subset of source data, generates annotator model data describing the characteristics of the at least one annotator device, and generates source data model data describing at least one piece of source data in the set of source data, where the source data model data includes label data identifying the estimated ground truth for each piece of source data in the set of source data.

Abstract: A pattern recognition apparatus for discriminative training includes: a similarity calculator that calculates similarities among training data; a statistics calculator that calculates statistics from the similarities in accordance with current labels for the training data; and a discriminative probabilistic linear discriminant analysis (PLDA) trainer that receives the training data, the statistics of the training data, the current labels and PLDA parameters, and updates the PLDA parameters and the labels of the training data.

Abstract: A text preparation apparatus is configured to in the decoding processing: perform first-layer recurrent neural network processing for phrase types to be used in the text and second-layer recurrent neural network processing for words appropriate for each of the phrase types; determine a phrase appropriate for each of the phrase types based on outputs of the second-layer recurrent neural network processing; generate a first vector set from a state vector of a previous step in the first-layer recurrent neural network processing and the feature vector sets, each vector of the first vector set being generated based on similarity degrees between individual vectors in one of the feature vector sets and the state vector; generate a second vector based on similarity degrees between individual vectors in the first vector set and the state vector; and input the second vector to a given step in the first-layer recurrent neural network processing.

Abstract: Examples of techniques for anomaly detection with feedback are described. An instance includes a technique is receiving a plurality of unlabeled data points from an input stream; performing anomaly detection on a point of the unlabeled data points using an anomaly detection engine; pre-processing the unlabeled data point that was subjected to anomaly detection; classifying the pre-processed unlabeled data point; determining the anomaly detection was not proper based on a comparison of a result of the anomaly detection and a result of the classifying of the pre-processed unlabeled data point; and in response to determining the anomaly detection was not proper, providing feedback to the anomaly detection engine to change at least one emphasis used in anomaly detection.

Abstract: Methods, systems, and apparatus, including computer programs encoded on a computer storage medium, for training a recurrent neural network on training sequences using backpropagation through time. In one aspect, a method includes receiving a training sequence including a respective input at each of a number of time steps; obtaining data defining an amount of memory allocated to storing forward propagation information for use during backpropagation; determining, from the number of time steps in the training sequence and from the amount of memory allocated to storing the forward propagation information, a training policy for processing the training sequence, wherein the training policy defines when to store forward propagation information during forward propagation of the training sequence; and training the recurrent neural network on the training sequence in accordance with the training policy.

Abstract: Computer-implemented systems and methods build ensembles for deep learning through parallel data splitting by creating and training an ensemble of up to 2n ensemble members based on a single base network and a selection of n network elements. The ensemble members are created by the “blasting” process, in which training data are selected for each of the up to 2n ensemble members such that each of the ensemble members trains with updates in a different direction from each of the other ensemble members. The ensemble members may also be trained with joint optimization.

Abstract: According to an embodiment of the present invention, a system designates each document in a collection of documents as a member of a first group containing known subjects for a concept of interest or as a member of a second group containing candidate subjects for the concept of interest and determines a subset of documents for at least one subject. The system generates a classifier based on the documents in the first and second groups and applies the classifier to a set of documents for the at least one subject to determine whether each document belong to the first and/or second group. The system generates a score for the at least one subject based on a quantity of documents for that subject assigned to the first group of documents relative to a total quantity of documents for that subject and ranks that subject based on the determined score for each subject.