Software and methods for controlling neural responses in deep brain regions
AssigneeMassachusetts Institute of Technology
InventorsJames Dicarlo, Pouya Bashivan, Kohitij Kar
Life of this patent
Estimating the term…
Abstract
Techniques for non-invasively controlling targeted neural activity of a subject are provided herein. The techniques include applying a stimulus input to the subject, the stimulus input being formed by a deep artificial neural network (ANN) model and being configured to elicit targeted neural activity within a brain of the subject. The stimulus input may be a pattern of luminous power generated by the deep ANN model and applied to retinae of the subject. The stimulus input may be generated by the deep ANN model based on a mapping of the subject's neural responses to neurons of the deep ANN model.
Claims20
A method of controlling targeted neural activity of a subject, the method comprising:
- applying a stimulus input to the subject, the stimulus input being formed by a deep artificial neural network (ANN) model and being configured to elicit targeted neural activity within a brain of the subject.
A system to control neural activity of a subject, the system comprising:
- at least one processor configured to access stimulus input images stored by at least one computer memory;
- and
- an apparatus, coupled to the processor, configured to apply the stimulus input images to retinae of the subject, wherein the stimulus input images are formed by a deep artificial neural network (ANN) model.
A method of controlling targeted neural activity of a subject, the method comprising:
- generating a stimulus input image configured to control targeted neural activity of the subject, the generating comprising:
- mapping neural activity of the subject responsive to one or more naturalistic images of a brain region to a deep artificial neural network (ANN) model;
- generating an initial image comprising random pixel values;
- masking the initial image with a receptive field of a neuron of the brain region of the subject to form a masked image;
- and
- using the deep ANN model to synthesize the stimulus input image based on the masked image;
- and
- applying the generated stimulus input image to retinae of the subject.
Dependencies are inferred from the claim text.