A search method for a neural network model structure, includes: generating an initial generation population of network model structure based on multi-objective optimization hyper parameters, as a current generation population of network model structure; performing selection and crossover on the current generation population of network model structure; generating a part of network model structure based on reinforcement learning mutation, and generating a remaining part of network model structure based on random mutation on the selected and crossed network model structure; generating a new population of network model structure based on the part of network model structure generated by reinforcement learning mutation and the remaining part of network model structure generated by random mutation; and searching a next generation population of network model structure based on the current generation population of network model structure and the new population of network model structure.

Techniques in advanced deep learning provide improvements in one or more of accuracy, performance, and energy efficiency, such as accuracy of learning, accuracy of prediction, speed of learning, performance of learning, and energy efficiency of learning. An array of processing elements performs flow-based computations on wavelets of data. Each processing element has a respective compute element and a respective routing element. Each compute element has processing resources and memory resources. Each router enables communication via wavelets with at least nearest neighbors in a 2D mesh. Stochastic gradient descent, mini-batch gradient descent, and continuous propagation gradient descent are techniques usable to train weights of a neural network modeled by the processing elements. Reverse checkpoint is usable to reduce memory usage during the training.

Techniques in advanced deep learning provide improvements in one or more of accuracy, performance, and energy efficiency, such as accuracy of learning, accuracy of prediction, speed of learning, performance of learning, and energy efficiency of learning. An array of processing elements performs flow-based computations on wavelets of data. Each processing element has a respective compute element and a respective routing element. Each compute element has processing resources and memory resources. Each router enables communication via wavelets with at least nearest neighbors in a 2D mesh. Stochastic gradient descent, mini-batch gradient descent, and continuous propagation gradient descent are techniques usable to train weights of a neural network modeled by the processing elements. Reverse checkpoint is usable to reduce memory usage during the training.

For machine training and application of a trained complex-valued machine learning model, an activation function of the machine learning model, such as a neural network, includes a learnable parameter that is complex or defined in a complex domain with two dimensions, such as real and imaginary or magnitude and phase dimensions. The complex learnable parameter is trained for any of various applications, such as MR fingerprinting, other medical imaging, or non-medical uses.

In certain embodiments, intent prediction and dialogue generation may be facilitated. In some embodiments, a chat initiation request may be obtained from a user. The latest activity information associated with the user may be provided to a prediction model to obtain a first set of predicted intents of the user. For each intent of the first set of predicted intents, a candidate question may be selected from a question set based on the candidate question matching the intent. In some embodiments, the candidate questions may be simultaneously presented on the chat interface.

In one or more implementations, the apparatus, systems and methods disclosed herein are directed to calibrating a smart phone with an arbitrary phone case for color lookup applications, wherein the calibration process includes obtaining, with the smartphone without the case equipped, a first measurement data set that includes at least one measurement of each of a black, white and grey calibration target; obtaining, with the smartphone with the case equipped, at least three exposure measurements of a white calibration target at least three different exposure times; calculating an optimized exposure time using at least the at least three exposure measurements; obtaining, with the smartphone with the case equipped, a second measurement data set that includes at least one measurement of each of a black, white and grey calibration target at the optimized exposure time; generating fitting parameters from the first and second measurement datasets; and storing the generated fitting parameters and optimized exposure time in at least one of a local or remote data storage device.

A dynamic graph includes a plurality of nodes and edges at a plurality of time steps; each node corresponds to a geographic location in a first area where pest infestation information is available for a subset of locations. Each edge connects two of the nodes which are geographically proximate, has a direction based on wind direction, and has a weight based on relative wind speed. Assign node features based on weather data as well as labels corresponding to pest infestation severity. Train a graph convolutional network on the dynamic graph. Based on predicted future weather conditions for a second area different than the first area, use the trained graph convolutional network to predict, via inductive learning, pest infestation severity for future times for a new set of nodes corresponding to new geographic locations in the second area for which no pest infestation information is available.

A method of multi-scale neural image compression with intra-prediction residuals is performed by at least one processor and includes downsampling an input image, generating a current predicted image, based on a previously-recovered predicted image, and generating a prediction residual based on a difference between the downsampled input image and the generated current predicted image. The method further includes encoding the generated prediction residual, decoding the encoded prediction residual, and generating a currently-recovered predicted image based on an addition of the current predicted image and the decoded prediction residual. The method further includes upsampling the currently-recovered predicted image, generating a scale residual based on a difference between the input image and the upsampled currently-recovered predicted image, and encoding the scale residual.

Briefly stated, the invention is directed to retrieving a semantically matched knowledge structure. A question and answer pair is received, wherein the answer is received from a query of a search engine. A question is constraint-matched with the answer based on maximizing a plurality of constraints, wherein at least one of the plurality of the constraints is a similarity score between question and answer, wherein the constraint matching generates a matched sequence. For one or more answer sequences, a subsequence is found that are not parsed as answer slots. Query results are obtained from another search engine based on a combination of the answer or question, and the non-answer subsequence. And a KB based is refined on the query results and the constraint matching and based on a neural network training, for a further subsequent semantic matching, wherein the KB includes a dense semantic vector indication of concepts.

According to one aspect of an embodiment a learning apparatus includes a first acquiring unit that acquires first output information that is output by an output layer when predetermined input information is input to a model that includes an input layer, a plurality of intermediate layers, and the output layer. The learning apparatus includes a second acquiring unit that acquires intermediate output information that is based on pieces of intermediate information that are output by the plurality of intermediate layers when the input information is input to the model. The learning apparatus includes a learning unit that learns the model based on the first output information and the intermediate output information.