Abstract: Methods, systems, and apparatus, including computer programs encoded on computer storage media, for generating object predictions using a neural network. One of the methods includes receiving respective projections of a plurality of channels of input sensor data, wherein each channel of input sensor data represents different respective characteristics of electromagnetic radiation reflected off of one or more objects. Each of the projections of the plurality of channels of input sensor data are provided to a neural network subsystem trained to receive projections of input sensor data as input and to provide an object prediction as an output. At the output of the neural network subsystem, an object prediction that predicts a region of space that is likely to be occupied by an object is received.

Abstract: A method is provided for determining at least one candidate reactant. One embodiment of this method includes the following steps: forming by a computer processor a graph of known reactants and known products, the graph comprising links between the known reactants and their known products, receiving by a computer processor the target compound, determining by a computer processor whether the graph includes the target compound and adding the target compound to the graph if the target compound was not previously included, forming by a computer processor a matrix representing at least a portion of the known reactants, a portion of the known products and the target compound, providing a matrix value of the graph by a computer processor for one or more candidate reactants and determining by a computer processor at least one link in the graph between the target compound and the candidate reactant based on matrix values.

Abstract: A machine learning device performs reinforcement learning on a controller that performs multiple processes for controlling a machine tool in parallel at multiple operation units.

Abstract: Described is a system for pattern recognition designed for neuromorphic hardware. The system generates a spike train of neuron spikes for training patterns with each excitatory neuron in an excitatory layer, where each training pattern belongs to a pattern class. A spiking rate distribution of excitatory neurons is generated for each pattern class. Each spiking rate distribution of excitatory neurons is normalized, and a class template is generated for each pattern class from the normalized spiking rate distributions. An unlabeled input pattern is classified using the class templates. A mechanical component of an autonomous device can be controlled based on classification of the unlabeled input pattern.

Abstract: Hardware accelerator for compressed Long Short Term Memory (LSTM) is disclosed. The accelerator comprise a sparse matrix-vector multiplication module for performing multiplication operation between all sparse matrices in the LSTM and vectors to sequentially obtain a plurality of sparse matrix-vector multiplication results. A addition tree module are also included for accumulating a plurality of said sparse matrix multiplication results to obtain an accumulated result. And a non-linear operation module passes the accumulated results through an activation function to generate non-linear operation result. That is, the present accelerator adopts pipeline design to overlap the time of data transfer and computation for compressed LSTM.

Abstract: A plurality of processing entities of a processor complex is maintained, wherein each processing entity has a local cache and the processor complex has a shared cache and a shared memory. One of the plurality of processing entities is allocated for execution of a critical task. In response to the allocating of one of the plurality of processing entities for the execution of the critical task, other processing entities of the plurality of processing entities are folded. The critical task utilizes the local cache of the other processing entities that are folded, the shared memory, and the shared cache, in addition to the local cache of the processing entity allocated for the execution of the critical task.

Abstract: Described is a system for storing and retrieving a memory in context. A memory formed for a given context is encoded in a neural model of the entorhinal-hippocampal system, forming a context-appropriate memory. The context-appropriate memory is comprised of an association between presented environmental cues and presence of a rewarded event. The system is able to discriminate between environmental cues in an environment surrounding a vehicle and retrieve at least one encoded context-appropriate memory. Using the at least one retrieved encoded context-appropriate memory, the system determines whether to initiate a collision avoidance operation to cause the vehicle to proactively avoid a collision.

Abstract: For a platform device (100) located at a web-site and capable of forming a network with a plurality of ID-detectable users or participants for gathering and processing items of information stored in each case in code-identifiable storage spaces (110) of a platform (100) assigned to a plurality of different topics by the users or participants, a passive automated distribution of information collected by the users or participants is enable in that the storage spaces (110) on the platform (100) are in each case formed by a dual unit DuU, a dual unit DuU comprising in each case a first storage space (111) that is assigned to a predefinable topic and provided with initial items of information (114) formulated by an initial participant and is not editable by any other individual user or participant and also a second storage space (112) assigned to the first storage space (111) and editable by any individual user or participant and construed for insertion, by the plurality of users or participants, of additional inf

Abstract: Embodiments include identifying unusual activity in an IT system based on user configurable message anomaly scoring. Aspects include receiving a message stream for the IT system and dividing the message stream into a plurality of intervals, wherein each interval corresponds to a time period. Aspects also include identifying and removing one or more intervals from the plurality of intervals that include a startup or a shutdown of an element of the IT system, identifying and removing one or more intervals from the plurality of intervals that correspond to a standard level of command activity and an elevated level of user complaint activity, and identifying and removing one or more intervals from the plurality of intervals that correspond to an elevated level of command activity and an standard level of user complaint activity. Aspects further include creating a training set of intervals that consists of the remaining labelled intervals.

Abstract: Methods, systems, and computer programs are provided for training a front-end neural network (“front-end NN”) and a back-end neural network (“back-end NN”). The method includes: combining the back-end NN with the front-end NN so that an output layer of the front-end NN is also an input layer of the back-end NN to form a joint layer to thereby generate a combined NN; and training the combined NN for a speech recognition with a set of utterances as training data, a plurality of specific units in the joint layer being dropped during the training and the plurality of the specific units corresponding to one or more common frequency bands. The front-end NN may be configured to estimate clean frequency filter bank features from noisy input features; or, to estimate clean frequency filter bank features from noisy frequency filter bank input features in the same feature space.

Abstract: A method for establishing sensorimotor programs includes specifying a concept relationship that relates a first concept to a second concept and establishes the second concept as higher-order than the first concept; training a first sensorimotor program to accomplish the first concept using a set of primitive actions; and training a second sensorimotor program to accomplish the second concept using the first sensorimotor program and the set of primitive actions.

Abstract: Various neural network models are constructed flexibly on a learning circuit. A machine learning model construction device includes a learning circuit (80) capable of constructed a neural network model according to a setting value, and a control means (11) capable of adjusting the setting value so as to become a value for constructing a predetermined neural network model in the learning circuit (80).

Abstract: In an example method, a mobile device receives a first calendar item associated with a first event. The first calendar item includes a first text string. The mobile device determines a correlation between the first text string and one or more locations associated with one or more second events. The mobile device determines a suggested location for the first event based on the correlation.

Abstract: Methods, systems, and computer program products for iterative and targeted feature selection are provided herein. A computer-implemented method includes generating a first prediction value for a variable attribute of a set of objects by executing a predictive model that comprises a set of features for the set of objects; evaluating the prediction error of the predictive model based on said first prediction value; generating additional features upon a determination that the prediction error exceeds a threshold; incorporating the additional features into the predictive model, generating an updated predictive model; generating a second prediction value for the variable attribute by executing the updated predictive model; evaluating the prediction error of the updated predictive model based on said second prediction value; and outputting the second prediction value to a user upon a determination that the prediction error of the updated predictive model is below the threshold.

Abstract: This disclosure involves personalizing user experiences with electronic content based on application usage data. For example, a user representation model that facilitates content recommendations is iteratively trained with action histories from a content manipulation application. Each iteration involves selecting, from an action history for a particular user, an action sequence including a target action. An initial output is computed in each iteration by applying a probability function to the selected action sequence and a user representation vector for the particular user. The user representation vector is adjusted to maximize an output that is generated by applying the probability function to the action sequence and the user representation vector. This iterative training process generates a user representation model, which includes a set of adjusted user representation vectors, that facilitates content recommendations corresponding to users' usage pattern in the content manipulation application.

Abstract: A system and method is provided for managing memory allocated to a virtual machine running on a host platform. An exemplary method includes continuously calculating the amount of free physical memory of the host platform by subtracting the amount of physical memory currently used consumed by the host operating system from the total size of the physical memory on the host platform. Moreover, using the calculated amount of free physical memory, the method includes dynamically adjusting an overall limit of the physical memory that can be allocated to the virtual machine running on the host platform, and then allocating to the virtual machine an amount this allocated physical memory so that active pages can be stored in the allocated memory and directly accessed during operation by the virtual machine.

Abstract: Aspects of the disclosure generally relate to computing devices and/or systems, and may be generally directed to devices, systems, methods, and/or applications for learning an avatar's or an application's operation in various circumstances, storing this knowledge in a knowledgebase (i.e. neural network, graph, sequences, etc.), and/or enabling autonomous operation of the avatar or the application.

Abstract: A mechanism is provided for modifying an existing recipe to meet a set of desired colors for a final food dish. Responsive to receiving a request to modify the existing recipe to meet the set of desired colors, at least one of the set of existing colors to be changed to meet the desired set of colors is identified. An ingredient-action-sequence triplet associated with each at least one existing color to be changed in the existing recipe is identified and, from a corpus of ingredient-action-sequence triplets associated with other existing recipes, one or more candidates that can be added to produce the at least one target color are identified. The one or more candidates that can be added are ranked based on how each candidate pairs best with other ingredients in the existing recipe. Based on a selection of a candidate, the existing recipe is modified with the selected candidate.

Abstract: A method and apparatus for utilizing virtual machines to pool memory from disparate server systems that may have disparate types of memory is described. The method may include establishing communication between a pool virtual machine and two or more publisher virtual machines. The method may also include aggregating, by the pool virtual machine, portions of memory from each of two or more publisher servers to generate a pool of memory, and providing an application with access to the pool of memory, through the pool virtual machine.

Abstract: A mechanism is provided in a data processing system for performing regression testing on a question answering system instance. The mechanism trains a machine learning model for a question answering system using a ground truth virtual checksum as part of a ground truth including domain-specific ground truth. The ground truth virtual checksum comprises a set of test questions, an answer to each test question, and a confidence level range for each answer to a corresponding test question. The mechanism runs regression test buckets across system nodes with domain-specific corpora and receiving results from the system nodes. Each system node implements a question answering system instance of the question answering system by executing in accordance with the machine learning model and by accessing domain-specific corpora. Each test bucket includes a set of questions matching a subset of questions in the ground truth virtual checksum.