Abstract: Provided are a neural network method and an apparatus, the method including obtaining a set of floating point data processed in a layer included in a neural network, determining a weighted entropy based on data values included in the set of floating point data, adjusting quantization levels assigned to the data values based on the weighted entropy, and quantizing the data values included in the set of floating point data in accordance with the adjusted quantization levels.

Abstract: A method for enhancing association rules includes: performing an association rule algorithm to establish a list of established association rules, wherein the list of established association rules includes at least one antecedent item set, at least one consequent item set and at least one original confidence; performing minimization of a cost function to obtain vector(s) of the at least one antecedent item set and vector(s) of the at least one consequent item set according to the list of established association rules, wherein the vector(s) of the at least one antecedent item set and the vector(s) of the at least one consequent item set correspond to the at least one antecedent item set and the at least one consequent item set; and establishing an enhanced association list according to the vector(s) of the at least one antecedent item set and the vector(s) of the at least one consequent item set.

Abstract: Some embodiments provide a method for compiling a neural network (NN) program for an NN inference circuit (NNIC) that includes multiple partial dot product computation circuits (PDPCCs) for computing dot products between weight values and input values. The method receives an NN definition with multiple nodes. The method assigns a group of filters to specific PDPCCs. Each filter is assigned to a different set of the PDPCCs. When a filter does not have enough weight values equal to zero for a first set of PDPCCs to which the filter is assigned to compute dot products for nodes that use the filter, the method divides the filter between the first set and a second set of PDPCCs. The method generates program instructions for instructing the NNIC to execute the NN by using the first and second PDPCCs to compute dot products for the nodes that use the filter.

Abstract: This disclosure relates to learning from distributed data. In particular, it relates to determining multiple first training samples from multiple first data samples. Each of the multiple first data samples comprises multiple first feature values and a first label that classifies that first data sample. A processor determines each of the multiple first training samples by selecting a first subset of the multiple first data samples such that the first subset comprises data samples with corresponding one or more of the multiple first feature values, and combining the first feature values of the data samples of the first subset based on the first label of each of the first data samples of the first subset. The resulting training samples can be combined with training samples from other databases that share the same corresponding features and entity matching is unnecessary.

Abstract: A method and system for classifying a vehicle based on low frequency GPS tracks. The method and system comprise retrieving a low frequency GPS track having a sampling interval of at least 20 seconds; deriving additional data from the low frequency GPS track, the additional data including interval speed and instantaneous acceleration of the vehicle; extracting a plurality of data sets from the low frequency GPS track and the additional data; generating a plurality of features from the extracted data sets; and providing the plurality of generated features to a classifier that classifies the vehicle into a predetermined class.

Abstract: A model optimizer is disclosed for managing training of models with automatic hyperparameter tuning. The model optimizer can perform a process including multiple steps. The steps can include receiving a model generation request, retrieving from a model storage a stored model and a stored hyperparameter value for the stored model, and provisioning computing resources with the stored model according to the stored hyperparameter value to generate a first trained model. The steps can further include provisioning the computing resources with the stored model according to a new hyperparameter value to generate a second trained model, determining a satisfaction of a termination condition, storing the second trained model and the new hyperparameter value in the model storage, and providing the second trained model in response to the model generation request.

Abstract: Systems and methods to infer or predict the proper placement of unstructured data (such as text, phrases, segments of phrases, alphanumeric characters) into a more structured format (such as a specific data field). In some embodiments, this is based on a user's prior assignment of similar unstructured data into a specific structure. In some embodiments, this may be based on other users' prior assignment of similar unstructured data into the specific structure. In yet other embodiments, this may be based on information obtained from business data used by a data processing platform to assist in operating the business (i.e., either business data or the output of a business application that processes the business data, such as an ERP, CRM, or eCommerce application).

Abstract: In general, embodiments of the present invention provide systems, methods and computer readable media for an adaptive oracle-trained learning framework for automatically building and maintaining models that are developed using machine learning algorithms. In embodiments, the framework leverages at least one oracle (e.g., a crowd) for automatic generation of high-quality training data to use in deriving a model. Once a model is trained, the framework monitors the performance of the model and, in embodiments, leverages active learning and the oracle to generate feedback about the changing data for modifying training data sets while maintaining data quality to enable incremental adaptation of the model.

Abstract: Methods, systems, and apparatus, including computer programs encoded on computer storage media, for augmenting neural networks with an external memory. One of the methods includes providing an output derived from a first portion of a neural network output as a system output; determining one or more sets of writing weights for each of a plurality of locations in an external memory; writing data defined by a third portion of the neural network output to the external memory in accordance with the sets of writing weights; determining one or more sets of reading weights for each of the plurality of locations in the external memory from a fourth portion of the neural network output; reading data from the external memory in accordance with the sets of reading weights; and combining the data read from the external memory with a next system input to generate the next neural network input.

Abstract: A mechanism is provided for text classifier training. The mechanism receives a training set of text and class specification pairs to be used as a ground truth for training a text classifier machine learning model for a text classifier. Each text and class specification pair comprises a text and a corresponding class specification. A domain terms selector component identifies at least one domain term in the texts of the training set. A domain terms replacer component replaces the at least one identified domain term in the texts of the training set with a corresponding replacement term to form a revised set of text and class specification pairs. A text classifier trainer component trains the text classifier machine learning model using the revised set to form a trained text classifier machine learning model.

Abstract: A method of updating a neural network may be provided. A method may include selecting a number of neurons for a layer for a neural network such that the number of neurons in the layer is less than at least one of a number of neurons in a first layer of the neural network and a number of neurons in a second, adjacent layer of the neural network. The method may further include and at least one of inserting the layer between the first layer and the second layer of the neural network and replacing one of the first layer and the second layer with the layer to reduce a number of connections in the neural network.

Abstract: A semantic embedding model using geometrical set-centric approach to capture both ABox and TBox representational models is disclosed. The model transforms a semantic-rich knowledge graph into a set of overlapping, disjoint, and/or subsumed n-dimensional spheres that captures and represents semantics embedded in the knowledge graph.

Abstract: Embodiments for fault diagnosis and analysis of refrigeration condenser systems by a processor. An energy usage anomaly is detected in a condenser by comparing an energy usage profile of the condenser against a knowledge domain of energy usage standards and energy usage standards anomalies.

Abstract: Some implementations of this specification are directed generally to deep machine learning methods and apparatus related to predicting motion(s) (if any) that will occur to object(s) in an environment of a robot in response to particular movement of the robot in the environment. Some implementations are directed to training a deep neural network model to predict at least one transformation (if any), of an image of a robot's environment, that will occur as a result of implementing at least a portion of a particular movement of the robot in the environment. The trained deep neural network model may predict the transformation based on input that includes the image and a group of robot movement parameters that define the portion of the particular movement.

Abstract: One factor in limiting the speed of conventional implementations of mixture models is that the algorithm involves many decisions where different operations are fetched and performed depending on the outcome of the decisions. These decisions cause flushing of the pipeline, and thus prevent the realization of a highly parallel pipeline in a processor. Without parallelism, the throughput of the pipeline in the processor, i.e., the ability to process many samples of the digital input at a time, is limited. To alleviate this issue, implementation of the mixture model is reformulated, among other things, by embedding decisions into the process flow as multiplicative factors. The resulting implementation alleviates the need to use if-else statements for the decisions and reduces the number of times the pipeline has to be flushed. The implementation enables a pipeline with a higher degree of parallelism and thereby increases throughput and speed of the implementation.

Abstract: A computer-implemented method for training a neural network system. The method includes receiving at least a first data vector at a first layer of the neural network system; applying a function to the first data vector to generate at least a second data vector, wherein the function is based on a layer parameter of the first layer that includes at least a weight matrix of the first layer; comparing at least the first data vector and the second data vector to obtain a loss value that represents a difference between the first data vector and the second data vector; updating the layer parameter based on the loss value; and adjusting the layer parameter based on a comparison of the updated layer parameter with a threshold value of the first layer.

Abstract: The present disclosure provides methods for applying artificial neural networks to flow cytometry data generated from biological samples to diagnose and characterize cancer in a subject. The disclosure also provides methods of training, testing, and validating artificial neural networks.

Abstract: Embodiments of the present invention provide an artificial neural network system for feature pattern extraction and output labeling. The system comprises a first spiking neural network and a second spiking neural network. The first spiking neural network is configured to autonomously learn complex, temporally overlapping features arising in an input pattern stream. Competitive learning is implemented as spike timing dependent plasticity with lateral inhibition in the first spiking neural network. The second spiking neural network is connected with the first spiking neural network through dynamic synapses, and is trained to interpret and label the output data of the first spiking neural network. Additionally, the labeled output of the second spiking neural network is transmitted to a computing device, such as a central processing unit for post processing.

Abstract: Methods, systems, and apparatus, including computer programs encoded on computer storage media, for augmenting neural networks with an external memory. One of the systems includes a sparse memory access subsystem that is configured to perform operations comprising generating a sparse set of reading weights that includes a respective reading weight for each of the plurality of locations in the external memory using the read key, reading data from the plurality of locations in the external memory in accordance with the sparse set of reading weights, generating a set of writing weights that includes a respective writing weight for each of the plurality of locations in the external memory, and writing the write vector to the plurality of locations in the external memory in accordance with the writing weights.

Abstract: A system is configured to receive first training data, train a first neural network (NN) based on the first training data, receive second training data, train a second NN based on the second training data, receive a first plain English phrase, provide the first plain English phrase to the first NN, generate, via the first NN, one or more first legal clauses based on the first plain English phrase, receive a second plain English phrase, provide the second plain English phrase to the first NN, generate, via the first NN, one or more second legal clauses based on the second plain English phrase, provide the one or more first legal clauses and the one or more second legal clauses to the second NN, and generate, via the second NN, a legal document based on the one or more first legal clauses and the one or more second legal clauses.