Abstract: A method of generating a controller for a continuous process. The method includes receiving from a storage memory off-line stored values of one or more controlled variables and one or more manipulated variables of the continuous process over a plurality of time points. The off-line stored values are used to train a first neural network to operate as a predictor of the controlled variables. Then, the method includes training a second neural network to operate as a controller of the continuous process using the first neural network after it was trained to operate as the predictor for the continuous process and employing the second neural network as a controller of the continuous process.

Abstract: Certain embodiments may generally relate to various techniques for machine learning. Feed-forward, fully-connected Artificial Neural Networks (ANNs), or the so-called Multi-Layer Perceptrons (MLPs) are well-known universal approximators. However, their learning performance may vary significantly depending on the function or the solution space that they attempt to approximate for learning. This is because they are based on a loose and crude model of the biological neurons promising only a linear transformation followed by a nonlinear activation function. Therefore, while they learn very well those problems with a monotonous, relatively simple and linearly separable solution space, they may entirely fail to do so when the solution space is highly nonlinear and complex. In order to address this drawback and also to accomplish a more generalized model of biological neurons and learning systems, Generalized Operational Perceptrons (GOPs) may be formed and they may encapsulate many linear and nonlinear operators.

Abstract: Methods and systems for reinforcement learning with dynamic agent grouping include gathering information at a first agent using one or more sensors. Shared information is received at the first agent from a second agent. An agent model is trained at the first agent using the gathered information and the shared information. A contribution of the shared information is weighted according to a degree of similarity between the first agent and the second agent. An action is generated using the trained agent model responsive to the gathered information.

Abstract: Systems and methods for lean parsing are disclosed. An example method is performed by one or more processors of a system and includes retrieving form data including first sentence segments and second sentence segments, determining a first predicate structure for each of the sentence segments based on a set of operators within the first set of sentence segments, identifying known tokens within the second set of sentence segments, each of the known tokens appearing on a list of predetermined tokens, identifying new tokens within the second set of sentence segments, each of the new tokens not on the list, mapping each known and new token to at least one operator, determining a second predicate structure for each sentence segment based on the mapping, and generating a predicate argument structure incorporating the first and second predicate structures, the predicate argument structure ready for mapping to at least one machine executable function.

Abstract: A method for processing information by an intelligent agent and the intelligent agent, where the method comprises: a first intelligent agent sends a request message to a second intelligent agent, where the request message includes an invitation message or a recommendation message; the first intelligent agent receives a decision message fed back by the second intelligent agent, where the decision message is determined according to the invitation message or the recommendation message and a knowledge model of the second intelligent agent; and the first intelligent agent updates, according to the decision message, a knowledge model of the first intelligent agent or sends a notification message to a first user account corresponding to the first intelligent agent. By using these technical solutions, information on a social network may be learned and processed by means of interaction with another intelligent agent, thereby implementing mining of data on the social network.

Abstract: Audio processing may be performed with a convolutional neural network that includes positional embeddings. Audio data may be received at an audio processing system. A convolutional neural network that concatenates frequency-positional embeddings at an input layer may be used to process the audio data. A result of processing the audio data through the convolutional neural network may be used to perform an audio processing task.

Abstract: A method may involve, for each of one or more messages that are selected from a plurality of messages from an account: (a) extracting one or more phrases from a respective selected message; (b) determining that a conversation includes the respective selected message and one or more other messages from the plurality of messages; (c) generating a first feature vector based on the conversation, wherein the first feature vector includes one or more first features, wherein the one or more first features include one or more words from the conversation; and (d) generating, by a computing system, one or more training-data sets, wherein each training-data set comprises one of the phrases and the first feature vector.

Abstract: Methods and systems are provided for neural architecture search. A computer-implemented neural architecture search may be used for providing a neural network configured to perform a selected task. A computational graph is obtained, which includes a plurality of nodes, edges and weightings associated with the nodes and/or edges. The computational graph includes a plurality of candidate models in the form of subgraphs of the computational graph. Selected subgraphs may be trained sequentially, with the weightings corresponding to each said subgraph being updated in response to training. For each weighting in a subgraph which is shared with another subgraph, updates to the weightings are controlled based on an indication of how important to another subgraph a node/edge associated with that weighting is.

Abstract: The present disclosure relates to a feature contribution system that accurately and efficiently provides the influence of features utilized in machine-learning models with respect to observed model results. In particular, the feature contribution system can utilize an observed model result, initial contribution values, and historical feature values to determine a contribution value correction factor. Further, the feature contribution system can apply the correction factor to the initial contribution values to determine correction-factor adjusted contribution values of each feature of the model with respect to the observed model result.

Abstract: The invention relates to a method and a system for determining the steel-tapping quantity of a converter, which consider that the working environment of the steel-making process of the converter is severe, the measurement is difficult and the interference of other factors is large, and provide a data-driven prediction model based on data, combine a Principal Component Analysis (PCA) with a RBF neural network, find the relation and the internal relation among variables by carrying out mathematical analysis on the related internal structure of the original variables, can quickly and accurately realize the prediction of the steel-tapping quantity of the converter, improve the component hit rate and the product stability in the steel-making process of the converter, are beneficial to realizing the control of narrow regions of steel-making components, save the alloying cost and have good application prospects in the field of ferrous metallurgy.

Abstract: Aspects of the invention include systems and methods to obtain meta features of a dataset for training in a deep learning application. A method includes selecting an initial search space that defines a type of deep learning architecture representation that specifies hyperparameters for two or more neural network architectures. The method also includes applying a search strategy to the initial search space. One of the two or more neural network architectures are selected based on a result of an evaluation according to the search strategy. A new search space is generated with new hyperparameters using an evolutionary algorithm and a mutation type that defines one or more changes in the hyperparameters specified by the initial search space, and, based on the mutation type, the new hyperparameters are applied to the one of the two or more neural networks or the search strategy is applied to the new search space.

Abstract: A technique for generating a performance prediction of a machine learning model with uncertainty intervals includes obtaining a first model configured to perform a task and a production dataset. At least one metric predicting a performance of the first model at performing the task on the production dataset is generated using a second model. The second model is a meta-model associated with the first model. At least one value predicting an uncertainty of the at least one metric predicting the performance of the first model at performing the task on the production dataset is generated using a third model. The third model is a meta-meta-model associated with the second model. An indication of the at least one metric and the at least one value is provided.

Abstract: Techniques related to electronic meeting intelligence are disclosed. An apparatus receives audio/video data including first meeting content data for an electronic meeting that includes multiple participants. The apparatus extracts the first meeting content data from the audio/video data. The apparatus generates meeting content metadata based on analyzing the first meeting content data. The apparatus includes the meeting content metadata in a report of the electronic meeting. If the apparatus determines that the audio/video data includes a cue for the apparatus to intervene in the electronic meeting, the apparatus generates intervention data including second meeting content data that is different from the first meeting content data. During the electronic meeting, the apparatus sends the intervention data to one or more nodes associated with at least one participant of the multiple participants.

Abstract: A neuromorphic device for the analog computation of a linear combination of input signals, for use, for example, in an artificial neuron. The neuromorphic device provides non-volatile programming of the weights, and fast evaluation and programming, and is suitable for fabrication at high density as part of a plurality of neuromorphic devices. The neuromorphic device is implemented as a vertical stack of flash-like cells with a common control gate contact and individually contacted source-drain (SD) regions. The vertical stacking of the cells enables efficient use of layout resources.

Abstract: This application describes systems and methods for generating machine learning models (MLMs). An exemplary method includes obtaining a sample and user input data characterizing a product or service. A subset of the data is selected from the sample based on sampling the sample according to the user input data. An MLM is trained by applying the data subset as training input to the MLM, thereby providing a trained MLM to emulate a customer selection process unique to the product or service. A user interface (UI) configured to receive other user input data and cause the trained MLM to execute on the other user input data, thereby testing the trained MLM, is presented. A summary of results from the execution of the trained MLM is generated and presented in the UI. The summary of results indicates a contribution to the trained MLM of each of a plurality of features.

Abstract: A classification device receives sensor data from a set of sensors and generates, using a context classifier having a set of classifier model parameters, a set of raw predictions based on the received sensor data. Temporal filtering and heuristic filtering are applied to the raw predictions, producing filtered predictions. A prediction error is generated from the filtered predictions, and model parameters of the set of classifier model parameters are updated based on said prediction error. The classification device may be a wearable device.

Abstract: A computer-implemented or hardware-implemented method of entity identification, comprising: a) providing a network of nodes with input from a plurality of sensors; b) generating, by each node of the network, an activity level, based on the input from the plurality of sensors; c) comparing the activity level of each node to a threshold level; d) based on the comparing, for each node, setting the activity level to a preset value or keeping the generated activity level; e) calculating a total activity level as the sum of all activity levels of the nodes of the network; f) iterating a)-e) until a local minimum of the total activity level has been reached; and g) when the local minimum of the total activity level has been reached, utilizing a distribution of activity levels at the local minimum to identify a measurable characteristic of the entity. The disclosure further relates to a computer program product and an apparatus for entity identification.

Abstract: The present disclosure discloses a system and method for distributed neural network training. The method includes: computing, by a plurality of heterogeneous computation units (HCUs) in a neural network processing system, a first plurality of gradients from a first plurality of samples; aggregating the first plurality of gradients to generate an aggregated gradient; computing, by the plurality of HCUs, a second plurality of gradients from a second plurality of samples; aggregating, at each of the plurality of HCUs, the aggregated gradient with a corresponding gradient of the second plurality of gradients to generate a local gradient update; and updating, at each of the plurality of HCUs, a local copy of a neural network with the local gradient update.

Abstract: A method for establishing a data-recognition model includes: generating (Z) number of Y-combinations of dithering algorithms from (X) number of dithering algorithms; for each Y-combination, performing a dithering operation on a to-be-processed data group, so as to obtain, in total, (Z) number of size-reduced data groups; performing training operations on a deep neural network using the size-reduced data groups, respectively, so as to generate, for each training operation, a DNN model and a steady deviation; and selecting the Y-combination corresponding to the size-reduced data group that results in the smallest steady deviation as a filter module, and selecting the corresponding DNN model as the data-recognition model.

Abstract: A method includes dividing a plurality of filters in a first layer of a neural network into a first set of filters and a second set of filters, applying each of the first set of filters to an input of the neural network, aggregating, at a second layer of the neural network, a respective one of a first set of outputs with a respective one of a second set of outputs, splitting respective weights of specific neurons activated in each remaining layer, at each specific neuron from each remaining layer, applying a respective filter associated with each specific neuron and a first corresponding weight, obtaining a second set of neuron outputs, for each specific neuron, aggregating one of the first set of neuron outputs with one of a second set of neuron outputs and generating an output of the neural network based on the aggregated neuron outputs.