Exponentially Decaying Memory for Neural Signal Classification
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Implantable medical devices, such as neural devices, face challenges in storing and classifying long-term neural signals due to limited memory capacity, leading to loss of signal characteristics and inefficiencies in power consumption.
Innovation Solution
A computer-implemented method and system using a machine learning model with exponentially decaying memory (EDM) to classify time series data by defining continuous sampling windows with exponential decay rates, allowing for efficient storage and classification of long-term trends without large memory requirements, utilizing processors and memory units to execute training and classification tasks.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Quantity of substance
If window-based approaches are used to classify neural signals, then classification can be performed with limited memory, but long-term signal characteristics are lost
Solution Approach 1:
The patent replaces the traditional mechanical window-based approach with a reservoir computing system that uses recurrent neural networks to process continuous data streams. This substitution allows the system to maintain long-term signal characteristics without requiring large memory capacity, as the RNN inherently captures temporal dependencies through its recurrent connections.
Solution Approach 2:
The patent changes the parameter of data representation from fixed-size windows to continuously updated feature vectors that incorporate exponential decay weighting. This parameter change allows the system to retain long-term information while adapting to recent changes, resolving the contradiction between memory limitations and information retention.
2Loss of information
If large memory is used to store long-term neural data, then complete signal characteristics are preserved, but device footprint and power consumption increase
Solution Approach 1:
The patent extracts only the essential temporal features from the neural signal stream using a reservoir computing approach. Instead of storing complete raw data, the system extracts and maintains compressed feature representations that capture the necessary signal characteristics, thereby reducing memory requirements and associated power consumption while preserving critical information.
Solution Approach 2:
The patent creates a virtual model of the neural signal processing through a reservoir computing system that simulates complex temporal relationships without requiring physical storage of the entire signal history. This virtual copying approach allows the system to access long-term signal characteristics on demand without the energy cost of storing and managing large amounts of raw data.
3Measurement precision
If more data is stored for long-term trend analysis, then classification accuracy improves, but memory requirements exceed device capacity
Solution Approach 1:
The patent implements a dynamic feature extraction system where the reservoir computing model continuously adapts its internal state to reflect changing signal characteristics. This dynamic approach allows the system to maintain high classification accuracy by processing data in a continuous stream with adaptive weighting, eliminating the need to store large static datasets while preserving the ability to detect long-term trends.
Data Source
AI summary
There is provided a system and method for classifying time series data for state identification. The method including: training a machine learning model to classify occurrences of the state; receiving a new time series data stream; determining whether a current sample in the new time series data stream is an occurrence of the state by determining a classified feature vector, the classified feature vector determined by passing the current sample and samples in at least one continuous sampling window into the trained machine learning model, each continuous sampling window including a plurality of preceding samples from the time series data, an epoch for each respective continuous sampling window determined according to a respective exponential decay rate; and outputting the determination of whether the current sample is an occurrence of the state.