Abstract: In one embodiment, an embedding is determined for each entity in a set of entities that is selected from a plurality of entities. Each embedding corresponds to a point in an embedding space, which includes points corresponding to embeddings of entities. The embeddings of the entities are determined using a deep-learning model. Embeddings are determined for each entity attribute in a set of entity attributes. Each of the entity attributes in the set is of an entity-attribute type and is associated with at least one entity. The entity-attribute embeddings are refined using the deep-learning model. The embeddings of the entities in the set of entities are modified based on the entity-attribute embeddings that are associated with the respective entity to obtain updated embeddings for each entity in the set. The updated embeddings include information regarding the entity attributes that are associated with the respective entities.

Abstract: Systems and methods for monitoring and assessing crop health and performance can provide rapid screening of individual plants. The systems and methods have an automated component, and rely primarily on the detection and interpretation of plant-based signals to provide information about crop health. In some cases knowledge from human experts is captured and integrated into the automated crop monitoring systems and methods. Predictive models can also be developed and used to predict future health of plants in a crop.

Abstract: An example method of designing an electrical machine includes providing at least one goal and at least one design constraint for a desired electrical machine to a deep neural network that comprises a plurality of nodes representing a plurality of prior electrical machine designs, the plurality of nodes connected by weights, each weight representing a correlation strength between two nodes. A proposed design is generated from the deep neural network for an electrical machine based on the goal(s) and design constraint(s). A plurality of the weights are adjusted based on a reward that rates at least one aspect of the proposed design. The proposed design is modified using the deep neural network after the weight adjustment. The adjusting and modifying are iteratively repeated to generate subsequent iterations of the proposed design, each subsequent iteration based on the reward from a preceding iteration. A system for designing electrical machines is also disclosed.

Abstract: A method of identifying a home to facilitate a real estate transaction includes generating a set of neural network objects wherein each object includes a neural network model, an address, and an image, training the neural network model in each object to emit a confidence score by a neural network analyzing the image to identify key features, receiving an example address and example image, selecting a set of trained neural network objects wherein each object in the set of trained neural network objects includes a trained neural network model, landmark address, and landmark image, generating a set of confidence scores by applying the example image to the trained neural network model in each object, generating based on the set of confidence scores a result set, transmitting the result set to a user, and displaying the result in the user device.

Abstract: Methods, systems, and apparatus for performing convolutional computations using an optical system. In some aspects computations for a neural network are performed in a digital domain using electronic circuitry, the neural network including a convolutional layer. Input data for the convolutional layer of the neural network is obtained, and a convolution or correlation computation on the input data in an analog domain using an optical correlator module is performed to generate an optical correlator module output. Based on the optical correlator module output, data is processed through additional layers of the neural network in the digital domain using the electronic circuitry.

Abstract: Embodiments relate to a neural processor circuit including neural engines, a buffer, and a kernel access circuit. The neural engines perform convolution operations on input data and kernel data to generate output data. The buffer is between the neural engines and a memory external to the neural processor circuit. The buffer stores input data for sending to the neural engines and output data received from the neural engines. The kernel access circuit receives one or more kernels from the memory external to the neural processor circuit. The neural processor circuit operates in one of multiple modes, at least one of which divides a convolution operation into multiple independent convolution operations for execution by the neural engines.

Abstract: A computer-implemented method for anomaly detection based on deep learning includes acquiring a plurality of records, each record having a corresponding number of attributes, identifying outliers in the plurality of records using labels generated from processing the plurality of records through an ensemble of different deep learning models, wherein an output of at least one model is used as an input to at least one other model and detecting anomalies in the plurality of records using a probabilistic classifier based on plurality of records and labels.

Abstract: A computing device trains a neural network machine learning model. A forward propagation of a first neural network is executed. A backward propagation of the first neural network is executed from a last layer to a last convolution layer of a plurality of convolutional layers to compute a gradient vector for first weight values of the last convolution layer using observation vectors. A discriminative localization map is computed for each observation vector with the gradient vector using a discriminative localization map function. A forward and a backward propagation of a second neural network is executed to compute a second weight value for each neuron of the second neural network using the discriminative localization map computed for each observation vector. A predefined number of iterations of the forward and the backward propagation of the second neural network is repeated.

Abstract: An approach is provided for generating parking restriction data using a machine learning model. The approach involves determining a plurality of classification features associated with a set of labeled road links. Each of the labeled road links is labeled with a parking restriction label that indicates a parking restriction status of said each of the labeled road links. The approach also involves training the machine learning model to classify an unlabeled road link of the geographic database using the plurality of classification features. The approach further involves determining the plurality of classification features for the unlabeled road link. The approach further involves processing the plurality of classification features for the unlabeled road link using the trained machine learning model to associate an assigned parking restriction label to the unlabeled road link. The approach further involves storing the assigned parking restriction label as the parking restriction data.

Abstract: In an approach for providing a self-learning framework for performance analysis using content-oriented analysis, a processor initiates a performance analysis of a dump on a thread. A processor presents time information and an associated location of the time information. A processor analyzes the time information by registering the time information into a knowledge base to debug errors in a computer program. Subsequent to a query for dump information, a processor displays the analyzed time information, based on the performance analysis.

Abstract: Probability density ratios may be estimated by a method including obtaining a first sample set including a plurality of first samples and a second sample set including a plurality of second samples, wherein each of the first samples and the second samples is represented as a vector including a plurality of parameters, constructing at least one decision tree estimating a ratio of probability density p(x)/q(x) based on the first sample set and the second sample set, wherein p(x) is a probability density of the first samples corresponding to an input vector x and q(x) is a probability density of the second samples corresponding to the input vector x.

Abstract: By way of example, the technology disclosed by this document may be implemented in a method that includes receiving stored sensor data describing characteristics of a vehicle in motion at a past time and extracting features for prediction and features for recognition from the stored sensor data. The features for prediction may be input into a prediction network, which may generate a predicted label for a past driver action based on the features for prediction. The features for recognition may be input into a recognition network, which may generate a recognized label for the past driver action based on the features for recognition. In some instances, the method may include training prediction network weights of the prediction network using the recognized label and the predicted label.

Abstract: Probability density ratios may be estimated by a method including obtaining a first sample set including a plurality of first samples and a second sample set including a plurality of second samples, wherein each of the first samples and the second samples is represented as a vector including a plurality of parameters, constructing at least one decision tree estimating a ratio of probability density p(x)/q(x) based on the first sample set and the second sample set, wherein p(x) is a probability density of the first samples corresponding to an input vector x and q(x) is a probability density of the second samples corresponding to the input vector x.

Abstract: Techniques for generating dynamic dust emission risk index values via construction and use of a dynamic dust emission risk index model are provided. In one example, a computer-implemented method comprises generating, by a system operatively coupled to a processor, a dynamic dust emission risk index value based on a dynamic dust emission risk index model. The computer implemented method also includes supplying, by the system, the dynamic dust emission risk index value to an air quality model. Additionally, the computer implemented method further comprises generating, by the system, a dust emission forecast based on the air quality model.

Abstract: Systems and techniques for interest matched interaction initialization are described herein. A first set of profile attributes for a first user and second set of profile attributes for a second user may be obtained. A first set of data sources and a second set of data sources may be identified respectively using the first set and second set of profile attributes. A first dataset and a second dataset may be collected respectively using the first and second set of data sources. An interest vector model may be generated using data elements of the first dataset based on an interest identified in the first dataset. The second dataset may be evaluated using the interest vector model to identify the interest as a shared interest in the second dataset. An interaction initialization item may be generated by identifying content associated with the shared interest. The interaction initialization item may be transmitted to a device.

Abstract: A computer-implemented method for analyzing a first neural network via a second neural network according to a differentiable function. The method includes adding a derivative node to the first neural network that receives derivatives associated with a node of the first neural network. The derivative node is connected to the second neural network such that the second neural network can receive the derivatives from the derivative node. The method further includes feeding forward activations in the first neural network for a data item, back propagating a selected differentiable function, providing the derivatives from the derivative node to the second neural network as data, feeding forward the derivatives from the derivative node through the second neural network, and then back propagating a secondary objective through both neural networks. In various aspects, the learned parameters of one or both of the neural networks can be updated according to the back propagation calculations.

Abstract: An overall gradient vector is computed at a server from a set of ISA vectors corresponding to a set of worker machines. An ISA vector of a worker machine including ISA instructions corresponding to a set of gradients, each gradient corresponding to a weight of a node of a neural network being distributedly trained in the worker machine. A set of register values is optimized for use in an approximation computation with an opcode to produce an x-th approximate gradient of an x-th gradient. A server ISA vector is constructed in which a server ISA instruction in an x-th position corresponds to the x-th gradient in the overall gradient vector. A processor at the worker machine is caused to update a set of weights of the neural network, using the set of optimized register values and the server ISA vector, thereby completing one iteration of training.

Abstract: Systems and Methods for training a neural network represented as a computational graph are disclosed. An example method begins with obtaining data representing a computational graph. The computational graph is then augmented to generate a training computational graph for training the neural network using a machine learning training algorithm that includes computing a gradient of an objective function with respect to each of the parameters of the neural network. Augmenting the computational graph includes inserting a plurality of gradient nodes and training edges into the computational graph to generate a backward path through the computational graph that represents operations for computing the gradients of the objective function with respect to the parameters of the neural network. The neural network is trained using the machine learning training algorithm by executing the training computational graph.

Abstract: Graphical interactive model selection is provided. A dataset includes observation vectors defined for each value of a plurality of values of a group variable. A nonlinear model is trained with each plurality of observation vectors to describe the response variable based on the explanatory variable for each value of the plurality of values of the group variable. Nonlinear model results are presented within a first sub-window of a first window. An indicator of a request to perform parameter analysis of the nonlinear model results is received. A linear model is trained. Trained linear model results from the trained linear model are presented within a second sub-window of the first window for each parameter variable of the nonlinear model. Predicted response variable values are presented as a function of the explanatory variable and the factor variable value using the trained nonlinear model within a third sub-window of the first window.

Abstract: The present invention discloses a method and a system for filtering goods review information. The method comprises: acquiring a plurality of predetermined advertisement spam samples, each advertisement spam sample comprising a review text and a user identification; establishing an advertisement spam user identification library comprising the user identifications of the plurality of advertisement spam samples; and acquiring a new review comprising a user identification and a review text, and determining the new review as an advertisement spam review if the user identification of the new review is included in the advertisement spam user identification library. An advertisement spam review is identified according to a user identification that publishes the review in the present invention. A new method is provided in the technical field of identifying an advertisement spam review for solving the problem that messy spam reviews are difficult to identify.