Abstract: Techniques for learning and using content type embeddings. The content type embeddings have the useful property that a distance in an embedding space between two content type embeddings corresponds to a semantic similarity between the two content types represented by the two content type embeddings. The closer the distance in the space, the more the two content types are semantically similar. The farther the distance in the space, the less the two content types are semantically similar. The learned content type embeddings can be used in a content suggestion system as machine learning features to improve content suggestions to end-users.

Abstract: In some embodiments, an application receives a request to execute a convolutional neural network model. The application determines the computational complexity requirement for the neural network based on the computing resource available on the device. The application further determines the architecture of the convolutional neural network model by determining the locations of down-sampling layers within the convolutional neural network model based on the computational complexity requirement. The application reconfigures the architecture of the convolutional neural network model by moving the down-sampling layers to the determined locations and executes the convolutional neural network model to generate output results.

Abstract: A method, system and computer program product, the method comprising: creating a model representing underperforming cases; from a case collection having a total performance, and which comprises for each of a multiplicity of records: a value for each feature from a collection of features, a ground truth label and a prediction of a machine learning (ML) engine, obtaining one or more features; dividing the records into groups, based on values of the features in each record; for one group of the groups, calculating a performance parameter of the ML engine over the portion of the records associated with the group; subject to the performance parameter of the group being below the total performance in at least a predetermined threshold: determining a characteristic for the group; adding the characteristic of the group to the model; and providing the model to a user, thus indicating under-performing parts of the test collection.

Abstract: A learning mechanism with partially-labeled web images is provided while correcting the noise labels during the learning. Specifically, the mechanism employs a momentum prototype that represents common characteristics of a specific class. One training objective is to minimize the difference between the normalized embedding of a training image sample and the momentum prototype of the corresponding class. Meanwhile, during the training process, the momentum prototype is used to generate a pseudo label for the training image sample, which can then be used to identify and remove out of distribution (OOD) samples to correct the noisy labels from the original partially-labeled training images. The momentum prototype for each class is in turn constantly updated based on the embeddings of new training samples and their pseudo labels.

Abstract: Embodiments of this application provide a neural network model deployment method, a prediction method and a device. The described features can implement deployment of a neural network model to improve the universality of the deployment of the neural network model to the terminal device by obtaining a layer definition and an operation parameter of each network layer of an initial neural network model, executing a target network layer corresponding to the network layers, applying relational connections amongst the target network layers using a net class, converting the operation parameters into a preset format, obtaining a target operation parameter based on the preset format, loading a corresponding target operation parameter in the target network layer, and obtaining a target neural network model based on the target operation parameter.

Abstract: In various implementations, provided are systems and methods for operating a neural network that includes conditional structures. In some implementations, an integrated circuit can compute a result using a set of intermediate results, where the intermediate results are computed from the outputs of a hidden layer of the neural network. The integrated circuit can further test the result against a condition. The outcome of the test can determine a next layer that the integrated circuit is to execute, or can be used to determine that further execution of the neural network can be terminated.

Abstract: Embodiments for identifying stochastic models representing the individual decision makers in a computing environment by a processor. One or more non-deterministic (stochastic, probabilistic) models may be identified according to a sequence of outcomes from decisions of each of a plurality of decision makers.

Abstract: The present disclosure discloses an address information feature extraction method based on a deep neural network model. The present disclosure uses a deep neural network architecture, and transforms tasks, such as text feature extraction, address standardization construction and semantic-geospatial fusion, into quantifiable deep neural network model construction and training optimization problems. Taking a character in an address as a basic input unit, the address language model is designed to express it in vectors, then a key technology of standardization construction of Chinese addresses is realized through neural network target tasks.

Abstract: A method optimizes a training of a machine learning system. A conflict detection system discovers a conflict between a first training data and a second training data for a machine learning system, where the first training data and the second training data are ground truths that describe a same type of entity, and where the first training data and the second training data have different labels. In response to discovering the conflict between the first training data and the second training data for the machine learning system, an oracle adjusts the different labels of the first training data and the second training data. The machine learning system is then trained using the first training data and the second training data with the adjusted labels.

Abstract: A system and method for recommending demand-supply agent pairs for transactions uses a deep neural network on data of demand agents to produce a demand agent vector, which is used to select supply agents based on their likelihood of future transaction and to find k nearest neighbor demand agents for each of the demand agents. The candidate supply agents and the k nearest neighbor demand agents are then combined to produce candidate demand-supply agent pairs, which are used to find recommended demand-supply agent pairs by applying modeling using machine learning.

Abstract: Methods, systems, and apparatus, including computer programs encoded on a computer storage medium, for jointly training an encoder neural network and a decoder neural network. In one aspect, a method comprises: updating current values of a set of encoder parameters and current values of a set of decoder parameters using gradients of a reconstruction loss function that measures an error in a reconstruction of multi-modal data from a training example, wherein: the reconstruction loss function comprises a plurality of scaling factors that each scale a respective term in the reconstruction loss function that measures an error in the reconstruction of a corresponding proper subset of feature dimensions of the multi-modal data from the training example.

Abstract: The present disclosure is directed to methods and systems for knowledge distillation.

Abstract: A system for program synthesis using annotations based on enumeration patterns includes a memory device for storing program code, and at least one processor device operatively coupled to the memory device. The at least one processor device is configured to execute program code stored on the memory device to obtain a set of annotated terms including one or more terms each annotated with an enumeration pattern, translate problem text into a formal specification using natural language processing, the formal specification being described as a set of rules associated with predicates, and synthesize one or more terms satisfying the set of rules of the formal specification based on the set of annotated terms to generate a computer program.

Abstract: Embodiments of the present disclosure relate to splitting input data into smaller units for loading into a data buffer and neural engines in a neural processor circuit for performing neural network operations. The input data of a large size is split into slices and each slice is again split into tiles. The tile is uploaded from an external source to a data buffer inside the neural processor circuit but outside the neural engines. Each tile is again split into work units sized for storing in an input buffer circuit inside each neural engine. The input data stored in the data buffer and the input buffer circuit is reused by the neural engines to reduce re-fetching of input data. Operations of splitting the input data are performed at various components of the neural processor circuit under the management of rasterizers provided in these components.

Abstract: Methods, systems, and apparatus, including computer programs encoded on a computer storage medium, for generating multi-modal data archetypes. In one aspect, a method comprises obtaining a plurality of training examples, wherein each training example corresponds to a respective patient and includes multi-modal data, having a plurality of feature dimensions, that characterizes the patient; jointly training an encoder neural network and a decoder neural network on the plurality of training examples; and generating a plurality of multi-modal data archetypes that each correspond to a respective dimension of a latent space, comprising, for each multi-modal data archetype: processing a predefined embedding that represents the corresponding dimension of the latent space using the decoder neural network to generate multi-modal data, having the plurality of feature dimensions, that defines the multi-modal data archetype.

Abstract: One embodiment can provide a system for detecting anomaly for high-dimensional sensor data associated with one or more machines. During operation, the system can obtain sensor data from a set of sensor associated with one or machines, generate a first set of outputs by using a set of clustering models learned in parallel from the unlabeled sensor data and user-provided partial label information, generate a second set of outputs by using a set of feed-forward neural network (FNN) models learned in parallel from the first set of outputs and the unlabeled sensor data, and determine whether an anomaly is present in the operation of the one or more machines based on the second set of outputs and a user-specified threshold.

Abstract: A computer trains a neural network. A neural network is executed with a weight vector to compute a gradient vector using a batch of observation vectors. Eigenvalues are computed from a Hessian approximation matrix, a regularization parameter value is computed using the gradient vector, the eigenvalues, and a step-size value, a search direction vector is computed using the eigenvalues, the gradient vector, the Hessian approximation matrix, and the regularization parameter value, a reduction ratio value is computed, an updated weight vector is computed from the weight vector, a learning rate value, and the search direction vector or the gradient vector based on the computed reduction ratio value, and an updated Hessian approximation matrix is computed from the Hessian approximation matrix, the predefined learning rate value, and the search direction vector or the gradient vector based on the reduction ratio value. The step-size value is updated using the search direction vector.

Abstract: Certain aspects involve updating data structures to indicate relationships between attribute trends and response variables used for training automated modeling systems. For example, a data structure stores data for training an automated modeling algorithm. The training data includes attribute values for multiple entities over a time period. A computing system generates, for each entity, at least one trend attribute that is a function of a respective time series of attribute values. The computing system modifies the data structure to include the generated trend attributes and updates the training data to include trend attribute values for the trend attributes. The computing system trains the automated modeling algorithm with the trend attribute values from the data structure. In some aspects, trend attributes are generated by applying a frequency transform to a time series of attribute values and selecting, as trend attributes, some of the coefficients generated by the frequency transform.

Abstract: Techniques for detecting state changes in a system may include receiving a first neural network that is trained to detect when the system transitions into a first resulting state, wherein the system transitions into at least a first intermediate state prior to transitioning into the final resulting state; training the first neural network using a first plurality of inputs denoting the system in the first intermediate state; obtaining a plurality of sets of internal state information of the first neural network, each set of the plurality of sets denoting an internal state of the first neural network at a different point in time after the first neural network has processed at least a portion of the first plurality of inputs; and training a second neural network, using the plurality of sets of internal state information, to detect the first intermediate state.

Abstract: Methods, systems, and computer-readable storage media for determining, by an automated regression detection system (ARDS), that training of a ML model is complete, the ML model being a version of a previously trained ML model, and in response, automatically, by the ARDS: retrieving the ML model, executing regression testing and detection using the ML model, generating regression results relative to the previously trained ML model, and publishing the regression results.