Pseudoinverse Estimation for Selective Neuron Stimulation
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current methods for designing stimuli to treat neurological disorders are limited by the need for extensive data and reliance on model-based or trial-and-error approaches, making it difficult to efficiently estimate pseudoinverses and explore a large parameter space for selective neurostimulation.
Innovation Solution
A novel pseudoinverse estimation system that adapts regression techniques to directly estimate pseudoinverses, jointly learning a restricted domain and the inverse mapping, allowing for data-efficient design of selective electrical waveforms without relying on computational models, and an adaptive algorithm that iteratively refines the waveform design based on collected data.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Adaptability or versatility
If traditional model-based approaches are used for stimulus design, then the methodology is reliable and interpretable, but the practical applicability is limited and exploration of parameter space is restricted
Solution Approach 1:
The patent replaces traditional model-based stimulus design with a data-driven approach using neural networks to estimate pseudoinverses. This substitution enables exploration of larger parameter spaces without relying on explicit computational models, while maintaining practical applicability through direct data-driven waveform generation
Solution Approach 2:
The patent changes the fundamental parameters of stimulus design by moving from model-based parameter estimation to data-driven pseudoinverse estimation. This allows the system to handle complex, non-invertible forward mappings by directly learning inverse relationships from data, enabling novel stimulus designs that were previously inaccessible
2Adaptability or versatility
If data-driven approaches are used to estimate pseudoinverses, then the exploration of parameter space is enabled, but the data requirements become extensive and collection becomes expensive
Solution Approach 1:
The patent applies preliminary action by pre-computing and storing lookup tables of neural responses for various stimulus parameters during the training phase. This allows the system to efficiently query and retrieve appropriate stimuli during application without requiring extensive real-time data collection, significantly reducing operational data requirements
Solution Approach 2:
The patent creates a data-efficient approach by learning compressed representations of stimulus-response relationships through neural networks. The model captures essential patterns from limited training data and generalizes to new parameter combinations, effectively copying the essential information needed for stimulus design without requiring exhaustive data coverage
3Measurement precision
If the forward mapping is inverted to obtain stimulus parameters, then the desired neural response can be achieved, but the mapping is non-invertible in most cases requiring pseudoinverse estimation
Solution Approach 1:
The patent replaces complex mathematical pseudoinverse estimation with a data-driven neural network approach. The neural network learns the inverse mapping directly from data, handling the non-invertibility issue automatically through its training process, which simplifies the overall system architecture while maintaining precise neural response control
Solution Approach 2:
The patent introduces a neural network as an intermediary between the desired neural response and the stimulus parameters. This intermediary learns to map responses to stimuli by analyzing training data, effectively bridging the gap created by non-invertible forward mappings without requiring explicit mathematical inversion
Data Source
AI summary
Disclosed herein is a novel pseudoinverse estimation system and method that adapts regression techniques to directly estimate one or more pseudo inverses of a neuromodulation pathway, thereby circumventing the need of inverting an estimated forward model to design an electrical waveform that elicits a desired neural response. This is accomplished by the learning of a restricted domain that restricts the potential stimuli required to produce the desired neural response. Also disclosed herein is an adaptive, data-driven method providing a more selective design of an electrical waveform to elicit the desired neural response.