Spiking Recurrent Model Training for Retinal Prostheses

Resolve Bottlenecks,
Find Innovative SolutionsGenerate Solutions

Solution Overview

Problem

Current retinal prostheses for treating age-related macular degeneration (AMD) and retinitis pigmentosa (RP) either simplify images or use convolutional neural networks that consume high energy and lack biological similarity, making them unsuitable for effective visual perception.

Innovation Solution

A spiking recurrent model training method that predicts spike responses of retinal ganglion cells using a Poisson loss function and time backpropagation, reducing power consumption and improving biological similarity, is implemented in a retinal prosthesis.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Measurement precision

If a convolutional neural network is used to predict retinal ganglion cell responses, then prediction capability is achieved, but energy consumption increases and biological similarity decreases

Engineering Contradiction:

Improveprediction accuracyVSAvoidenergy consumption

Core Design Contradiction:

Measurement precisionVSUse of energy by moving object

Solution Approach 1:

The patent replaces the conventional convolutional neural network (which uses floating-point multiplication) with a spiking neural network model that mimics biological neuron behavior. This substitution uses spike-based signal processing instead of traditional mathematical operations, significantly reducing energy consumption while maintaining prediction capability for retinal ganglion cell responses.

Inventive Principle:

Principle #28Mechanics substitution (Replace mechanical system)

Solution Approach 2:

The patent changes the fundamental parameters of the neural network model from continuous floating-point values to discrete spike events. By transforming the representation and processing of neural signals to match biological spiking behavior, the system achieves both energy efficiency and improved biological similarity while preserving predictive accuracy.

Inventive Principle:

Principle #35Parameter changes

2Measurement precision

If a convolutional neural network is used to predict retinal ganglion cell responses, then prediction capability is achieved, but biological similarity decreases

Engineering Contradiction:

Improveprediction accuracyVSAvoidbiological similarity

Core Design Contradiction:

Measurement precisionVSAdaptability or versatility

Solution Approach 1:

The patent replaces the artificial mathematical framework of convolutional neural networks with a biologically-inspired spiking neural network framework. This substitution introduces spike-based communication, temporal coding, and event-driven processing that closely resemble actual retinal ganglion cell behavior, thereby improving biological similarity while maintaining prediction accuracy.

Inventive Principle:

Principle #28Mechanics substitution (Replace mechanical system)

Solution Approach 2:

The patent transforms the parameter representation from continuous floating-point values to discrete temporal spike events, matching the natural encoding mechanism of biological neurons. This parameter transformation enables the model to capture temporal dynamics and spike-based information processing characteristic of real retinal ganglion cells.

Inventive Principle:

Principle #35Parameter changes

3Productivity

If floating-point multiplication is used for processing, then computation is performed, but computation complexity and energy consumption increase

Engineering Contradiction:

Improvecomputation capabilityVSAvoidcomputation complexity

Core Design Contradiction:

ProductivityVSDevice complexity

Solution Approach 1:

The patent substitutes floating-point multiplication operations with spike-based event processing. Instead of performing continuous mathematical computations, the system uses discrete spike events that propagate through the network, eliminating the need for complex floating-point arithmetic hardware and reducing computational complexity.

Inventive Principle:

Principle #28Mechanics substitution (Replace mechanical system)

Solution Approach 2:

The patent employs event-driven, periodic spike processing rather than continuous computation. Spikes occur at specific moments based on input changes, allowing the system to perform computations only when necessary, thereby reducing overall computational complexity and energy consumption while maintaining productivity.

Inventive Principle:

Principle #19Periodic action

Data Source

PatentUS20240358895A1Model training method, visual perception method, electronic device and storage medium

Publication Date: 2024.10.31 WESTLAKE UNIV

AI summary

Disclosed are a model training method, a visual perception method, an electronic device and a storage medium. The model training method is applicable to a spiking recurrent model in a retinal prosthesis and includes: determining labels respectively corresponding to ganglion cells based on a preset ganglion cell response dataset; obtaining multiple spike signals as training samples; inputting the spike signals into the spiking recurrent model, obtaining spike responses of the ganglion cells predicted by the spiking recurrent model, and computing a loss value based on the spike responses of the ganglion cells, the labels and a preset Poisson loss function; and updating a weight for each layer in the spiking recurrent model based on the loss value and a preset time backpropagation function recursively until the spiking recurrent model converges, thereby scientifically and quickly training the spiking recurrent model for predicting the responses of retinal ganglion cells.