Qualcomm Incorporated

Intel Corporation

William Howard Constable

Venkat Rangan

Manu Rastogi

Xin Wang

Vikram Makam Gupta

Jeffrey Alexander Levin

Ravindra Manohar Patwardhan

Amir Khosrowshahi

Carey Kevin Kloss

Sarah Paige Gibson

William Richard Bell, Ii

Avijit Chakraborty

Scott Gray

Urs Koster

Luke James Hornof

A system receives and executes a sequence of tensor instructions, for example, instructions for performing a neural network computation. The system may be implemented as a multiprocessor architecture, for example, hardware for performing a neural network computation. A tensor instruction specifies a tensor computation receiving one or more input tensors for determining an output tensor. The system stores a decimal position associated with a plurality of values of a tensor. The system performs the tensor computation of a tensor instruction to determine a plurality of values of the output tensor. The system collects statistics describing the plurality of values of the output tensor and determines a decimal position for the plurality of values based on the collected statistics.

Dynamic management of numerical representation in a distributed matrix processor architecture

A method of interfacing an event based processing system with a frame based processing system is presented. The method includes converting multiple events into a frame. The events may be generated from an event sensor. The method also includes inputting the frame into the frame based processing system.

Interfacing an event based system with a frame based processing system

A method of event-based down sampling includes receiving multiple sensor events corresponding to addresses and time stamps. The method further includes spatially down sampling the addresses based on the time stamps and the addresses. The method may also include updating a pixel value for each of the multiple sensor events based on the down sampling.

Event-based down sampling

A method for computing a spatial Fourier transform for an event-based system includes receiving an asynchronous event output stream including one or more events from a sensor. The method further includes computing a discrete Fourier transform (DFT) matrix based on dimensions of the sensor. The method also includes computing an output based on the DFT matrix and applying the output to an event processor.

Event-based spatial transformation

A method for managing synapse plasticity in a neural network includes converting a first set of synapses from a plastic synapse type to a fixed synapse type. The method may also include converting a second set of synapses from the fixed synapse type to the plastic synapse type.

Plastic synapse management

Growing community of inventors

San Diego, CA, United States of America

William Howard Constable