Leland Stanford Junior University

Google Inc.

David Sussillo

Jonathan C Kao

Sergey Stavisky

Krishna V Shenoy

Jonathon Shlens

Niruban Maheswaranathan

A brain machine interface (BMI) to control a device is provided. The BMI has a neural decoder, which is a neural to kinematic mapping function with neural signals as input to the neural decoder and kinematics to control the device as output of the neural decoder. The neural decoder is based on a continuous-time multiplicative recurrent neural network, which has been trained as a neural to kinematic mapping function. An advantage of the invention is the robustness of the decoder to perturbations in the neural data; its performance degrades less—or not at all in some circumstances—in comparison to the current state decoders. These perturbations make the current use of BMI in a clinical setting extremely challenging. This invention helps to ameliorate this problem. The robustness of the neural decoder does not come at the cost of some performance, in fact an improvement in performance is observed.

Systems and methods for intracortical brain machine interface decoding

Multiplicative recurrent neural network for fast and robust intracortical brain machine interface decoders

A brain machine interface (BMI) to control a device is provided. The BMI has a neural decoder, which is a neural to kinematic mapping function with neural signals as input to the neural decoder and kinematics to control the device as output of the neural decoder. The neural decoder is based on a continuous-time multiplicative recurrent neural network, which has been trained as a neural to kinematic mapping function. An advantage of the invention is the robustness of the decoder to perturbations in the neural data; its performance degrades less&#x2014;or not at all in some circumstances&#x2014;in comparison to the current state decoders. These perturbations make the current use of BMI in a clinical setting extremely challenging. This invention helps to ameliorate this problem. The robustness of the neural decoder does not come at the cost of some performance, in fact an improvement in performance is observed.

A method for generating a segmentation of an image that assigns each pixel to a respective segmentation category from a set of segmentation categories is described. The method includes obtaining features of the image, the image including a plurality of pixels. For each of one or more time steps starting from an initial time step and continuing until a final time step, the method includes generating a network input from the features of the image and a current segmentation output as of the time step, processing the network input using a convolutional recurrent neural network to generate an intermediate segmentation output for the time step, and generating an updated segmentation output for the time step from the intermediate segmentation output for the time step and the current segmentation output as of the time step. The method includes generating a final segmentation of the image from the updated segmentation output.

Growing community of inventors

Portola Valley, CA, United States of America

David Sussillo