Abstract: A method and system for constructing a convolutional neural network (CNN) model are herein disclosed. The method includes regularizing spatial domain weights, providing quantization of the spatial domain weights, pruning small or zero weights in a spatial domain, fine-tuning a quantization codebook, compressing a quantization output from the quantization codebook, and decompressing the spatial domain weights and using either sparse spatial domain convolution and sparse Winograd convolution after pruning Winograd-domain weights.

Abstract: A machine learning system may be used to predict clinical questions to ask on a clinical form. A first encoder may encode first information and a second encoder may encoder second information from a medical record of a past appointment. The first and second encoded information and additional encoded information may be used to predict a clinical question to ask by using a reinforcement learning system. The reinforcement learning system may be trained by receiving ratings of questions from users.

Abstract: Systems and methods for generating reduced order models are provided herein. In embodiments, a set of learning points is identified in a parametric space. A 3D physical solver may be used to perform a simulation for each learning point in the set of learning points to generate a learning data set, where the 3D physical solver is selected from a plurality of compatible 3D physical solvers for simulating different physical aspects of a product or process. The learning data set may be compressed to reduce the learning data set to a smaller set of vectors. Coefficients from the learning data set and the smaller set of vectors may then be used to interpolate a set of coefficients within a design space for the reduced order model.

Abstract: Embodiments for training a neural network are provided. A neural network is divided into a first block and a second block, and the parameters in the first block and second block are trained in parallel. To train the parameters, a gradient from a gradient mini-batch included in training data is generated. A curvature-vector product from a curvature mini-batch included in the training data is also generated. The gradient and the curvature-vector product generate a conjugate gradient. The conjugate gradient is used to determine a change in parameters in the first block in parallel with a change in parameters in the second block. The curvature matrix in the curvature-vector product includes zero values when the terms correspond to parameters from different blocks.

Abstract: An optimization calculation method includes: generating, by a computer, current generation individuals with a selected previous generation individual as a parent individual; evaluating each current generation individual by using a predetermined evaluation function; calculating a current generation constraint condition value based on a previous generation constraint condition value and a constraint condition provisional value which is achieved by more than half of the current generation individuals; determining whether a result of the evaluation for each current generation individual satisfies the current generation constraint condition value; determining a predetermined offset based on an attribute of each individual, which is generated by a mutation generating process, among individuals having the evaluation results satisfying the current generation constraint condition value; and adding the predetermined offset to a random number used to generate each next generation individual by the mutation generating pr

Abstract: A multi-element problem may be solved iteratively by using a modified genetic algorithm to generate a plurality of solutions according to a set of solution criteria. The solution criteria may comprise a plurality of servers, each server including one or more attributes, and an indication of which of the one or more attributes are to be optimized. An index may be appended, by a processing device, to each solution in the plurality of solutions and the values in each solution may be sorted. For each solution in the plurality of solutions, one or more values from the solution may be combined with one or more values from another solution to generate a plurality of child solutions. Each child solution may have an index, and the values in each child solution may be sorted in view of the child solution's index. For one or more child solutions in the plurality of child solutions, two selected values may be rearranged to generate one or more mutated child solutions.

Abstract: A data analysis and processing method includes forming an initial assembly of datasets comprising multiple entities, where each entity is a collection of variables and relationships that define how entities interact with each other, simulating an evolution of the initial assembly by performing multiple iterations in which a first iteration uses the initial assembly as a starting assembly, and querying, during the simulating, the evolution of the initial assembly, for datasets that meet an optimality criterion.

Abstract: A computer-based method for simulating a human-like decision in an environmental context, comprising: capturing environmental data with at least one sensor, realising a computer based method for realising a bi-directional compression of high dimensional data by compressing the data into a lower-dimensional map, if environmental data are captured during a learning phase of a computer-based model, evaluate the map of compressed data by determining the quality of the map by how well it separates data with different properties, the captured data corresponding to known pre-recorded data that have been pre-evaluated, if environmental data are captured after the learning phase, add new point to the compressed data and generate a signal indicating which human-like decision to use to correspond to the state of the operator.

Abstract: In transactional systems where past transactions can have impact on the current score of a machine learning based decision model, the transactions that are most responsible for the score and the associated reasons are determined by the transactional system. A system and method identifies such past transactions that maximally impact the current score and allow for a more effective understanding of the scores generated by a model in a transactional system and explanation of specific transactions for automated decisioning, to explain the scores in terms of past transactions. Further an existing instance-based explanation system is used to identify the reasons for the score, and how the identified transactions influence these reasons. A combination of impact on score and impact on reasons determines the most impactful past transaction with respect to the most recent score being explained.

Abstract: An optimization calculation method includes, generating individuals of a present generation with an individual selected in a previous generation as a parent individual using an algorithm for obtaining optimum solutions while evolving a plurality of individuals for each generation, evaluating each individual of the present generation, setting a constraint condition value of the present generation based on a constraint condition value of one generation before with respect to the present generation, and a constraint condition provisional value achieved by a half or more of individuals used for generating child individuals of a next generation, determining whether an evaluation result of each individual satisfies a constraint condition value of the present generation, changing for lowering an evaluation result of individuals which do not satisfy the constraint condition value of the present generation, and selecting individuals to be solution candidates from among individuals on which the determination or change

Abstract: Techniques that include applying machine learning models to episode data, including a cardiac electrogram, stored by a medical device are disclosed. In some examples, based on the application of one or more machine learning models to the episode data, processing circuitry derives, for each of a plurality of arrhythmia type classifications, class activation data indicating varying likelihoods of the classification over a period of time associated with the episode. The processing circuitry may display a graph of the varying likelihoods of the arrhythmia type classifications over the period of time. In some examples, processing circuitry may use arrhythmia type likelihoods and depolarization likelihoods to identify depolarizations, e.g., QRS complexes, during the episode.

Abstract: A system for generating a data structure using graphical models includes a computing device configured to provide a visual interface configured to provide a plurality of graphical models of a plurality of rule modules and receive a relational identification of at least a graphical representation of the plurality of graphical models, the relational indication including at least an entry indication and at least an exit indication, to convert the relational identification into at least a decision tree having at least a root node corresponding and at least a terminal node, to train a machine-learning model to match execution parameters to the at least a root node, and to generate an execution result interface configured to receive at least an execution parameter, map it to the at least a root node using the machine-learning model, and generate an execution result at the at least a terminal node.

Abstract: In described examples, a quadrature phase shifter includes digitally programmable phase shifter networks for generating leading and lagging output signals in quadrature. The phase shifter networks include passive components for reactively inducing phase shifts, which need not consume active power. Output currents from the transistors coupled to the phase shifter networks are substantially in quadrature and can be made further accurate by adjusted by a weight function implemented using current steering elements. Example low-loss quadrature phase shifters described herein can be functionally integrated to provide low-power, low-noise up/down mixers, vector modulators and transceiver front-ends for millimeter wavelength (mmwave) communication systems.

Abstract: Embodiments relate to systematic explanation of neural model behavior and effective deduction of its vulnerabilities. Input data is received for the neural model and applied to the model to generate output data. Accuracy of the output data is evaluated with respect to the neural model, and one or more neural model vulnerabilities are identified that correspond to the output data accuracy. An explanation of the output data and the identified one or more vulnerabilities is generated, wherein the explanation serves as an indicator of alignment of the input data with the output data.

Abstract: In response to receiving observed data of mixed observed variables, a mixed causality objective function, being suitable for continuous observed variables and discrete observed variables is determined, wherein the mixed causality objective function includes a causality objective function for continuous observed variables and a causality objective function for discrete observed variables and the fitting inconsistency is adjusted based on weighted factors of the observed variables. Then, the mixed causality objective function is optimally solved by using a mixed sparse causal inference, suitable for both continuous observed variables and discrete observed variables, using the mixed observed data under a constraint of a directed acyclic graph, to estimate causality among the observed variables.

Abstract: A system comprising a dynamically configurable neural network engine operating on a plurality of processors, an automated knowledge matrix creation system operating on a processor and coupled to the dynamically configurable neural network engine, the automated knowledge matrix creation system configured to provide an input to the neural network engine, and a machine comprehension system coupled to the dynamically configurable neural network engine and configured to associate a plurality of product development documents with a customer created knowledge base and to provide input to the neural network engine.

Abstract: Computer-implemented methods, computer program products, and systems are provided for multi-step ahead forecasting. A method includes configuring, by a processor device, a Vector Autoregression (VAR) model to generate a multi-step-ahead forecast based on previous observations. The previous observations are predictors and the multi-step-ahead forecast is a response to the predictors. The method further includes training, by the processor device, the VAR model using complex-valued weight parameters to avoid a training result relating to any of a divergence and a convergence to zero.

Abstract: A model training method includes: acquiring a plurality of user data pairs, wherein data fields of two sets of user data in each user data pair have an identical part; acquiring a user similarity corresponding to each user data pair, wherein the user similarity is a similarity between users corresponding to the two sets of user data in each user data pair; determining, according to the user similarity corresponding to each user data pair and the plurality of user data pairs, sample data for training a preset classification model; and training the classification model based on the sample data to obtain a similarity classification model.

Abstract: Systems and associated methods are described for providing content recommendations. The system accesses a plurality of recommendation algorithms and assigns a plurality of weight values to each prediction algorithm. Then, the system generates a set of candidate weight combinations, such that each candidate combination includes a weight value assigned to each prediction algorithm. Then requests for content items are received over a predetermined period of time. For each combination, the system generates a set of recommended content items and an evaluation metric that is based on matches with requests. Afterwards, the system replaces a candidate combination that resulted in a generation of a lowest evaluation metric. The aforementioned steps are repeated until the evaluation metrics stop improving. Then display identifiers are displayed for a set of recommended content items generated for a candidate combination with the highest evaluation metric.

Abstract: Roughly described, a computer-implemented evolutionary system evolves candidate solutions to provided problems. It includes a memory storing a candidate gene database containing active and epigenetic individuals; a gene pool processor which tests only active individuals on training data and updates their fitness estimates; a competition module which selects active individuals for discarding in dependence upon both their updated fitness estimate and their testing experience level; and a gene harvesting module providing for deployment selected ones of the individuals from the gene pool. The gene database has an experience layered elitist pool, and individuals compete only with other individuals in their same layer. Certain individuals are made epigenetic in the procreation module, after which they are not subjected to testing and competition. Epigenetic individuals are retained in the candidate gene pool regardless of their fitness.