Phase-Coding Coordinate Transformation in Spiking Neural Networks
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current artificial neural networks face challenges in efficiently transforming allocentric coordinates to egocentric coordinates for applications like robot navigation, where traditional methods are cumbersome and impractical.
Innovation Solution
A method and apparatus for encoding positional representations as phase information in spiking neural networks, shifting this phase information to modify allocentric coordinates into egocentric coordinates, utilizing phase-coding techniques to adjust the spiking phase of neurons for coordinate transformation.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Ease of operation
If traditional computational techniques are used for coordinate transformation, then the transformation can be performed, but the process becomes cumbersome and impractical
Solution Approach 1:
The patent replaces traditional mechanical/computational coordinate transformation systems with a biological neural network system that performs the same function through biological mechanisms. The neural network uses spike timing and phase coding to transform allocentric coordinates to egocentric coordinates, substituting complex computational algorithms with biologically-inspired parallel processing that occurs naturally in the neural system.
2Productivity
If phase-coding techniques are applied to neural spikes, then coordinate transformation efficiency improves, but the system requires specialized neuromorphic architecture
Solution Approach 1:
The patent changes the parameter representation of spatial coordinates from traditional Cartesian coordinates to phase-coded temporal representations. By encoding spatial information in the phase of neural spikes rather than in static numerical values, the system achieves efficient coordinate transformation through temporal dynamics. This parameter transformation allows the neural network to perform coordinate transformations by modulating spike phases according to gain fields, converting spatial relationships into temporal relationships that can be processed biologically.
3Measurement precision
If gain field modulation is used for coordinate transformation, then the transformation accuracy improves, but the neural network requires precise phase control
Solution Approach 1:
The patent implements feedback mechanisms within the neural network to maintain and adjust phase relationships during coordinate transformation. The recurrent connections and lateral inhibition in the neural circuitry provide continuous feedback that stabilizes phase coding and ensures accurate gain field modulation. This feedback allows the system to self-correct phase deviations and maintain transformation accuracy without requiring external precision control, leveraging the inherent dynamics of the neural system.
Data Source
AI summary
A method for coordinate transformation in a spiking neural network includes encoding a first positional representation as phase information in the spiking neural network. The method also includes shifting the phase information to modify the first positional representation into a second positional representation.