On-line Neural Interface Calibration via Sparse REW-NPLS Regression

Resolve Bottlenecks,
Find Innovative SolutionsGenerate Solutions

Solution Overview

Problem

Existing direct neural interfaces face challenges in on-line calibration due to the need for extensive data manipulation and high computational requirements, particularly with methods like penalised multipath regression, which limits their ability to select relevant electrodes for efficient trajectory prediction and command signal calculation.

Innovation Solution

A method for on-line calibration of direct neural interfaces using a Recursive Exponentially Weighted N-way Partial Least Squares (REW-NPLS) regression with PARAFAC iterative decomposition and a penalisation term, which promotes sparsity in the prediction tensor by selectively penalising and thresholding projector elements, allowing for incremental calibration and efficient electrode selection.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Measurement precision

If penalised multipath regression is used for on-line calibration, then electrode selection and prediction accuracy are improved, but computational complexity and data manipulation requirements increase

Engineering Contradiction:

Improveprediction accuracyVSAvoidcomputational complexity

Core Design Contradiction:

Measurement precisionVSDevice complexity

Solution Approach 1:

The patent segments the calibration process into incremental updates rather than requiring complete recalibration. The predictive model is updated step-by-step using new data windows, dividing the complex calibration task into manageable incremental steps that reduce computational burden while maintaining accuracy.

Inventive Principle:

Principle #1Segmentation

Solution Approach 2:

The patent changes the parameter weighting scheme by introducing exponential weighting that decays over time. This allows the model to prioritize recent data while gradually reducing the influence of older data, improving prediction accuracy without requiring equal computational resources for all historical data points.

Inventive Principle:

Principle #35Parameter changes

2Reliability

If complete calibration data from multiple steps is accumulated, then prediction robustness is improved, but data manipulation complexity and processing time increase

Engineering Contradiction:

Improveprediction robustnessVSAvoidcalibration processing time

Core Design Contradiction:

ReliabilityVSLoss of time

Solution Approach 1:

The patent performs preliminary actions by maintaining a running predictive model that is continuously updated with new data. Rather than waiting to accumulate all calibration data before processing, the model is incrementally refined as each new data window becomes available, reducing overall processing time while maintaining robustness through continuous validation.

Inventive Principle:

Principle #10Preliminary action

Solution Approach 2:

The patent ensures continuity of useful action by implementing continuous incremental calibration rather than periodic batch processing. The model continuously adapts to new neural signals through ongoing updates, maintaining prediction robustness through constant refinement without the downtime associated with periodic complete recalibrations.

Inventive Principle:

Principle #20Continuity of useful action

3Adaptability or versatility

If on-line calibration is performed frequently, then model adaptability to non-stationary signals is improved, but computational resource consumption increases

Engineering Contradiction:

Improvemodel adaptabilityVSAvoidcomputational resource consumption

Core Design Contradiction:

Adaptability or versatilityVSUse of energy by moving object

Solution Approach 1:

The patent implements periodic action through a sliding window approach where calibration data is collected over fixed time windows and processed at regular intervals. This periodic processing rhythm allows the model to adapt to non-stationary signals at appropriate frequencies while avoiding excessive computational resource consumption by batching updates rather than processing continuously.

Inventive Principle:

Principle #19Periodic action

Solution Approach 2:

The patent applies dynamics by making the calibration update frequency adaptive rather than fixed. The system dynamically adjusts the balance between frequent updates for adaptability and spaced updates for resource conservation, allowing the model to respond to changing signal characteristics while managing computational resources based on actual needs rather than rigid schedules.

Inventive Principle:

Principle #15Dynamics

Data Source

PatentUS12059258B2Method for calibrating on-line and with forgetting factor a direct neural interface with penalised multivariate regression

Publication Date: 2024.08.13 COMMISSARIAT A LENERGIE ATOMIQUE ET AUX ENERGIES ALTERNATIVES

AI summary

The present invention relates to a method for calibrating on-line a direct neural interface implementing a REW-NPLS regression between an output calibration tensor and an input calibration tensor. The REW-NPLS regression comprises a PARAFAC iterative decomposition of the cross covariance tensor between the input calibration tensor and the output calibration tensor, each PARAFAC iteration comprising a sequence of M elementary steps (2401, 2401, . . . 240M) of minimisation of a metric according to the alternating least squares method, each elementary minimisation step relating to a projector and considering the others as constant, said metric comprising a penalisation term that is a function of the norm of this projector, the elements of this projector not being subjected to a penalisation during a PARAFAC iteration f not being penalisable during following PARAFAC iterations. Said calibration method makes it possible to obtain a predictive model of which the non-zero coefficients are sparse blockwise.