Abstract: An artificial neuron including: a membrane capacitor; an input of an external synaptic excitation in current, the membrane capacitor integrating the input current; a negative-feedback impulse circuit, supplied by a power supply at a negative voltage between ?200 mV and 0 mV and at a positive voltage between 0 mV and +200 mV, including: a bridge based on pMOS and nMOS transistors in series and linked by a midpoint to the membrane capacitor, the midpoint defining the output of the artificial neuron, at least one delay capacitor between the gate and the source of one of the transistors of the bridge, at least two CMOS inverters between the membrane capacitor and the gates of the transistors of the bridge.

Abstract: Some embodiments of the invention provide a novel method for training a quantized machine-trained network. Some embodiments provide a method of scaling a feature map of a pre-trained floating-point neural network in order to match the range of output values provided by quantized activations in a quantized neural network. A quantization function is modified, in some embodiments, to be differentiable to fix the mismatch between the loss function computed in forward propagation and the loss gradient used in backward propagation. Variational information bottleneck, in some embodiments, is incorporated to train the network to be insensitive to multiplicative noise applied to each channel. In some embodiments, channels that finish training with large noise, for example, exceeding 100%, are pruned.

Abstract: Techniques are disclosed for systems and methods for learning the behavior of and/or for performing automated testing of a system under test (SUT). The learning/testing is accomplished solely via the graphical user interface (GUI) of the SUT and requires no a priori metadata/knowledge about the GUI objects. The learning engine operates by performing actions on the GUI and by observing the results of these actions. If the actions result in a change in the screen/page of the GUI then a screenshot is taken for further processing. Objects are detected from the screenshot, new actions that need to be performed on the objects are guessed, those actions are performed, the results are observed and the process repeats. Text labels on the screens are also read and are used for generating contextualized inputs for the screens. The learning process continues until any predetermined learning/testing criteria are satisfied.

Abstract: Methods and apparatus for quantum assisted optimization. In one aspect, a method includes obtaining a set of initial input states, applying one or more of (i) dynamical thermal fluctuations and (ii) cluster update algorithms to the set of input states and subsequent input states when the states evolve within the classical information processors, applying dynamical quantum fluctuations to the set of input states and subsequent states when the states evolve within the quantum systems and repeating the application steps until a desirable output state is obtained.

Abstract: A system and method are shown for transferring weight information to analog non-volatile memory elements wherein the programming pulse duration is directly proportional to the difference in weights. Furthermore, the system and method avoid weight transfers when the weights are already well-matched.

Abstract: A method for pruning a neural network includes initializing a plurality of threshold values respectively corresponding to a plurality of layers included in the neural network; selecting one of the plurality of layers; adjusting the threshold value of the selected layer; and adjusting a plurality of weights respectively corresponding to a plurality of synapses included in the neural network.

Abstract: Systems, methods, and computer readable media related to determining one or more responses to provide that are responsive to an electronic communication that is generated through interaction with a client computing device. For example, determining one or more responses to provide for presentation to a user as suggestions for inclusion in a reply to an electronic communication sent to the user. Some implementations are related to training and/or using separate input and response neural network models for determining responses for electronic communications. The input neural network model and the response neural network model can be separate, but trained and/or used cooperatively.

Abstract: A method and associated systems provide real-time response to a request received from a real-time system like a self-driving vehicle or a device that communicates interactively with its environment. The response is selected from a set of candidate feasible responses by a group of computerized agents that each sort the feasible responses in order of that agent's specific preferences, based on that agent's particular priorities or expertise. The agents then reconcile their differences through an iterative procedure. During each iteration, each agent decides whether to retain its current preferences or to adopt the preferences of another agent. This decision is made by determining which preferences are most similar to that agent's own initial preferences, and by which preferences would be most useful in helping to achieve that agent's particular goals. When the agents reach consensus, the group's most-preferred response is returned quickly enough to provide real-time, interactive response.

Abstract: In one embodiment, a system to accelerate batch-normalized convolutional neural network (CNN) models is disclosed. The system extracts a plurality of first groups of layers from a first CNN model, each group of the first groups having a first convolutional layer and a first batch-norm layer. For each group of the plurality of first groups, the system calculates a first scale vector and a first shift vector based on the first batch-norm layer, and generates a second convolutional layer representing the corresponding group of the plurality of first groups based on the first convolutional layer and the first scale and the first shift vectors. The system generates an accelerated CNN model based on the second convolutional layer corresponding to the plurality of the first groups, such that the accelerated CNN model is utilized subsequently to classify an object perceived by an autonomous driving vehicle (ADV).

Abstract: Architectures for multicore neuromorphic systems are provided. In various embodiments, a neural network description is read. The neural network description describes a plurality of logical cores. A plurality of precedence relationships are determined among the plurality of logical cores. Based on the plurality of precedence relationships, a schedule is generated that assigns the plurality of logical cores to a plurality of physical cores at a plurality of time slices. Based on the schedule, the plurality of logical cores of the neural network description are executed on the plurality of physical cores.

Abstract: Disclosed herein are system, method, and computer program product embodiments for encoding symbolic data into a subsymbolic format while preserving the semantic arrangement of the symbolic data. In an embodiment, to encode the symbolic data, a subsymbolic encoder system may convert a symbolic graph into a tuple representation having tuple elements corresponding to the nodes of the symbolic graph. The subsymbolic encoder system may retrieve a dictionary identification for each tuple element and calculate a subsymbolic value for each tuple element using an exponential component. The subsymbolic encoder system may standardize the length of the subsymbolic values and/or add a weighted relationship indicator to the subsymbolic values. The subsymbolic encoder system may transmit the subsymbolic values to a subsymbolic intelligence system.

Abstract: Methods, systems, and apparatus for prioritizing communications are described. Metadata that characterizes an electronic communication is obtained and a machine learning algorithm is applied to the metadata to generate a scoring model. A score for the electronic communication is generated based on the scoring model.

Abstract: A method for selecting an action, includes reading, into a memory, a Partially Observed Markov Decision Process (POMDP) model, the POMDP model having top-k action IDs for each belief state, the top-k action IDs maximizing expected long-term cumulative rewards in each time-step, and k being an integer of two or more, in the execution-time process of the POMDP model, detecting a situation where an action identified by the best action ID among the top-k action IDs for a current belief state is unable to be selected due to a constraint, and selecting and executing an action identified by the second best action ID among the top-k action IDs for the current belief state in response to a detection of the situation. The top-k action IDs may be top-k alpha vectors, each of the top-k alpha vectors having an associated action, or identifiers of top-k actions associated with alpha vectors.

Abstract: Methods, systems, and apparatus, including computer programs encoded on computer storage media, for processing inputs using recurrent neural networks. One of the systems includes a main recurrent neural network comprising one or more recurrent neural network layers and a respective hyper recurrent neural network corresponding to each of the one or more recurrent neural network layers, wherein each hyper recurrent neural network is configured to, at each of a plurality of time steps: process the layer input at the time step to the corresponding recurrent neural network layer, the current layer hidden state of the corresponding recurrent neural network layer, and a current hypernetwork hidden state of the hyper recurrent neural network to generate an updated hypernetwork hidden state.

Abstract: A system for generating alerts including processors and storage devices. The instructions configure the one or more processors to perform operations, which include receiving an event from a data stream, extracting keys from the event, associating the event with at least one account based on the extracted keys, identifying a state variable associated with the at least one account, updating the state variable by accumulating the event in the state variable, registering a time stamp for the event in the state variable, and retiring expired events from the state variable. The operations may also include determining whether the state variable is above a threshold level and generating an alert for the account when the state variable is above the threshold level.

Abstract: A user interface may be presented to a creator to facilitate the creation of narrative content. The user interface may be part of a system configured to generate recommendations pertaining to narrative content. The narrative content is meant to be experienced by users, e.g., in a virtual space. Feedback and/or other responses from the creator may be used to train and/or modify the generation of new recommendations. Feedback and/or other responses from the users may be used to train and/or modify the generation of new recommendations.

Abstract: Methods, systems, and apparatus, including computer programs encoded on a computer storage medium, for performing operations using data from a data source. In one aspect, a method includes a neural network system including a controller neural network configured to: receive a controller input for a time step and process the controller input and a representation of a system input to generate: an operation score distribution that assigns a respective operation score to an operation and a data score distribution that assigns a respective data score in the data source. The neural network system can also include an operation subsystem configured to: perform operations to generate operation outputs, wherein at least one of the operations is performed on data in the data source, and combine the operation outputs in accordance with the operation score distribution and the data score distribution to generate a time step output for the time step.

Abstract: A system for remote monitoring of a machine is provided. The system includes a data store to store machine data associated with an operation of the machine. The system includes an analyzer comprising a plurality of analytics engines to analyze the machine data. The analyzer selects one or more analytics engines based at least on one of machine data and a type of the machine. The analyzer is further configured to analyze machine data using the selected one or more analytics engines and to determine a plurality of exceptions. The system includes a rules engine to process at least two of the plurality of exceptions and determine a smart exception, wherein the smart exception is a hierarchical combination of the at least two of the plurality of exceptions. The system includes an interface to display a notification to a user in the event of a smart exception.

Abstract: Techniques for generating a personality trait model are described. According to an example, a system is provided that can generate text data and linguistic data, and apply psycholinguistic data to the text data and the linguistic data, resulting in updated text data and updated linguistic data. The system is further operable to combine the updated text data with the updated linguistic data to generate a personality trait model. In various embodiments, the personality trait model can be trained and updated as additional data is received from various inputs.

Abstract: Methods, systems, and apparatus, including computer programs encoded on a computer storage medium, for training a machine learning model. In one aspect, a method includes receiving training data for training the machine learning model on a plurality of tasks, where each task includes multiple batches of training data. A task is selected in accordance with a current task selection policy. A batch of training data is selected from the selected task. The machine learning model is trained on the selected batch of training data to determine updated values of the model parameters. A learning progress measure that represents a progress of the training of the machine learning model as a result of training the machine learning model on the selected batch of training data is determined. The current task selection policy is updated using the learning progress measure.