Abstract: Embodiments of the present application disclose a method and apparatus for generating a neural network structure, an electronic device, and a storage medium. The method comprises: sampling a neural network structure to generate a network block, the network block comprising at least one network layer; constructing a sampling neural network based on the network block; training the sampling neural network based on sample data, and obtaining an accuracy corresponding to the sampling neural network; and in response to that the accuracy does not meet a preset condition, regenerating a new network block according to the accuracy until a sampling neural network constructed by the new network block meets the preset condition, and using the sampling neural network meeting the preset condition as a target neural network.

Abstract: A method for unsupervised learning of a multilevel hierarchical network of artificial neurons wherein each neuron is interconnected with artificial synapses to neurons of a lower hierarchical level and to neurons of an upper hierarchical level.

Abstract: A neural network learns a particular task by being shown many examples. In one scenario, a neural network may be trained to label an image, such as cat, dog, bicycle, chair, etc. In other scenario, a neural network may be trained to remove noise from videos or identify specific objects within images, such as human faces, bicycles, etc. Rather than training a complex neural network having a predetermined topology of features and interconnections between the features to learn the task, the topology of the neural network is modified as the neural network is trained for the task, eventually evolving to match the predetermined topology of the complex neural network. In the beginning the neural network learns large-scale details for the task (bicycles have two wheels) and later, as the neural network becomes more complex, learns smaller details (the wheels have spokes).

Abstract: A system for forecasting the drying of an agricultural crop includes an electronic processor configured to receive weather data associated with an agricultural field and receive an agricultural field parameter from a field sensor associated with the agricultural field. The electronic processor is also configured to determine a drying score for each of a plurality of harvest times based on the weather data and the agricultural field parameter. The electronic processor is also configured to determine a recommended harvest time for harvesting the agricultural crop based on the drying score, wherein the recommended harvest time is included in the plurality of harvest times. The electronic processor is also configured to output a forecast for the agricultural crop for display to a user, wherein the forecast includes the drying score and the recommended harvest time for harvesting the agricultural crop.

Abstract: An enhanced prediction model utilizing product life cycle segmentation and historic prediction data for generating more accurate future failure rates. Input data is segmented into groups based on failure modes of a corresponding life cycle. A prediction model such as Weibull analysis is implemented for each segmented group. Historical prediction data is also segmented into groups. Prediction parameters for each group of segmented historical prediction data are compared with one another and the comparisons are then used to adjust the prediction parameters generated from the segmented groups of input data. Updated parameters for the input data are then output thereby generating a new future failure rate.

Abstract: A generative adversarial neural network (GAN) learns a particular task by being shown many examples. In one scenario, a GAN may be trained to generate new images including specific objects, such as human faces, bicycles, etc. Rather than training a complex GAN having a predetermined topology of features and interconnections between the features to learn the task, the topology of the GAN is modified as the GAN is trained for the task. The topology of the GAN may be simple in the beginning and become more complex as the GAN learns during the training, eventually evolving to match the predetermined topology of the complex GAN. In the beginning the GAN learns large-scale details for the task (bicycles have two wheels) and later, as the GAN becomes more complex, learns smaller details (the wheels have spokes).

Abstract: A device may receive information that identifies a requirement. The device may receive information associated with a set of positive entities and a set of negative entities. The device may identify a set of priority terms based on the information associated with the set of positive entities and the set of negative entities. The device may identify a first set of ancillary terms and a second set of ancillary terms based on the information that identifies the set of priority terms. The device may generate a model based on the set of priority terms, the first set of ancillary terms, and the second set of ancillary terms. The device may determine a set of classification scores, for a set of unclassified entities, based on information associated with the set of unclassified entities and the model. The device may provide information that identifies the set of classification scores to cause an action to be performed in association with the set of unclassified entities.

Abstract: A circuit for a neuron of a multi-stage compute process is disclosed. The circuit comprises a weighted charge packet (WCP) generator. The circuit may also include a voltage divider controlled by a programmable resistance component (e.g., a memristor). The WCP generator may also include a current mirror controlled via the voltage divider and arrival of an input spike signal to the neuron. WCPs may be created to represent the multiply function of a multiply accumulate processor. The WCPs may be supplied to a capacitor to accumulate and represent the accumulate function. The value of the WCP may be controlled by the length of the spike in signal times the current supplied through the current mirror. Spikes may be asynchronous. Memristive components may be electrically isolated from input spike signals so their programmed conductance is not affected. Positive and negative spikes and WCPs for accumulation may be supported.

Abstract: A method, system and computer-usable medium are disclosed for automated analysis of ground truth using confidence model to prioritize correction options. In certain embodiments, the ground truth data is analyzed to identify review-candidates. A confidence level may be assigned to each of the identified review-candidates and the review-candidates are prioritized, at least in part, using the assigned confidence levels. The review-candidates are electronically presented in prioritized order to solicit verification or correction feedback for updating the ground truth data.

Abstract: Limited duration supply for heuristic algorithms is disclosed. A supply manager receives, from a first subsystem, a first request for a first supply. The supply manager determines that the first supply is not executing. The supply manager initiates the first supply, the first supply to return supply data upon request. The supply manager provides to the first subsystem a supply reference that refers to the first supply that allows the first subsystem to request the supply data directly from the first supply. The supply manager subsequently determines that no subsystem requires the first supply and disables the first supply.

Abstract: Methods, systems, and apparatus, including computer programs encoded on computer storage media, for training a document classification neural network. One of the methods includes training an autoencoder neural network to autoencode input documents, wherein the autoencoder neural network comprises the one or more LSTM neural network layers and an autoencoder output layer, and wherein training the autoencoder neural network comprises determining pre-trained values of the parameters of the one or more LSTM neural network layers from initial values of the parameters of the one or more LSTM neural network layers; and training the document classification neural network on a plurality of training documents to determine trained values of the parameters of the one or more LSTM neural network layers from the pre-trained values of the parameters of the one or more LSTM neural network layers.

Abstract: A score explanation method for explaining a score includes at least steps of: a1) providing a first score associated with a first vector containing the first values of the parameters; b) generating a first set of lists, each list including a second number of indicators; c) generating, from a list, of at least a third vector wherein each parameter has a third value; the third value being equal to the corresponding first value when the list does not include an indicator of the corresponding parameter, and different from the corresponding first value otherwise; d) calculating the score of at least one third vector; e) evaluating, from the scores calculated for each of the third vectors, an indicator of significance of each parameter; f) elaborating, from the evaluated indicators of significance, an explanation of the first score.

Abstract: Described herein is a crossbar array that includes a cross-point synaptic device at each of a plurality of crosspoints. The cross-point synaptic device includes a weight storage element comprising a set of nanocrystal dots. Further, the cross-point synaptic device includes at least three terminals for interacting with the weight storage element, wherein a weight is stored in the weight storage element by sending a first electric pulse via a gate terminal from the at least three terminals, the first electric pulse causes the nanocrystal dots to store a corresponding charge, and the weight is erased from the weight storage element by sending a second electric pulse via the gate terminal, the second electric pulse having an opposite polarity of the first electric pulse.

Abstract: Embodiments relate to a first processing node that processes an input data having a temporal sequence of spatial patterns by retaining a higher-level context of the temporal sequence. The first processing node performs temporal processing based at least on feedback inputs received from a second processing node. The first processing node determines whether learned temporal sequences are included in the input data based on sequence inputs transmitted within the same level of a hierarchy of processing nodes and the feedback inputs received from an upper level of the hierarchy of processing nodes.

Abstract: Systems and methods disclosed herein relate to autonomous agents. A first autonomous agent receives, from a first sensor, a first set of event data indicating events relating to a subject. The first autonomous agent provides the first set of event data to a data aggregator. The first autonomous agent receives, from the data aggregator, correlated event data including events sensed by the first autonomous agent and a second autonomous agent. The first autonomous agent applies machine learning model to the correlated event data to predict a first pattern of activity and determines, based on the first pattern of activity, that a first action is to be performed, causing the first actuator module to perform the first action.

Abstract: A computer-implemented method of mixed-precision deep learning with multi-memristive synapses may be provided. The method comprises representing, each synapse of an artificial neural network by a combination of a plurality of memristive devices, wherein each of the plurality of memristive devices of each of the synapses contributes to an overall synaptic weight with a related device significance, accumulating a weight gradient ?W for each synapse in a high-precision variable, and performing a weight update to one of the synapses using an arbitration scheme for selecting a respective memristive device, according to which a threshold value related to the high-precision variable for performing the weight update is set according to the device significance of the respective memristive device selected by the arbitration schema.

Abstract: A computer-implemented method for reducing computation cost associated with a machine learning task performed by a computer system by implementing continuous control of attention for a deep learning network includes initializing a control-value function, an observation-value function and a sequence of states associated with a current episode. If a current epoch associated with the current episode is odd, an observation-action is selected, the observation-action is executed to observe a partial image, and the observation-value function is updated based on the partial image and the control-value function. If the current epoch is even, a control-action is selected, the control-action is executed to obtain a reward corresponding to the control-action, and the control-value function is updated based on the reward and the observation-value function.

Abstract: A machine learning model inference routing system in a machine learning service is described herein. The machine learning model inference routing system includes load balancer(s), network traffic router(s), an endpoint registry, and a feedback processing system that collectively allow the machine learning model inference routing system to adjust the routing of inferences based on machine learning model accuracy, demand, and/or the like. In addition, the arrangement of components in the machine learning model inference routing system enables the machine learning service to perform shadow testing, support ensemble machine learning models, and/or improve existing machine learning models using feedback data.

Abstract: A trained computer model includes a direct network and an indirect network. The indirect network generates expected weights or an expected weight distribution for the nodes and layers of the direct network. These expected characteristics may be used to regularize training of the direct network weights and encourage the direct network weights towards those expected, or predicted by the indirect network. Alternatively, the expected weight distribution may be used to probabilistically predict the output of the direct network according to the likelihood of different weights or weight sets provided by the expected weight distribution. The output may be generated by sampling weight sets from the distribution and evaluating the sampled weight sets.

Abstract: An approach is provided in which an information handing system determines a hidden cycle of hidden evidence based on one of multiple signals in a frequency-based representation of source evidence. The information handling system extrapolates the hidden evidence to create a forecast data set and, in turn, utilizes the forecast data set to process a request.