Mirrored Neuron Pairs for Robust Spiking Dynamics
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current neuromorphic systems face challenges in understanding how individual neuron spikes contribute to overall system objectives and how populations of neurons self-optimize to produce emergent spiking or population responses, such as noise-shaping, especially in large-scale implementations.
Innovation Solution
The growth transform neural network system introduces mirrored neuron pairs interconnected by a normalization link, with a growth transform module that updates each neuron pair based on a growth transform neuron model, and a network convergence module that solves a system objective function to achieve steady-state conditions, producing synchronized or spiking outputs.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Device complexity
If a simpler neuron model is used for large-scale implementations, then device complexity is reduced, but the ability to capture spiking dynamics and achieve robust recognition performance deteriorates
Solution Approach 1:
The patent uses mirrored neuron pairs where each pair consists of two neurons with symmetric connections. This copying approach allows the system to maintain spiking dynamics and robust recognition performance through the mirrored structure while using simpler individual neuron models. The mirrored pairs work together to capture temporal patterns and spiking behaviors that would require complex single-neuron models.
Solution Approach 2:
The patent divides the neural network into mirrored neuron pairs, segmenting the computational task across multiple simpler units rather than relying on a single complex neuron. Each mirrored pair processes information independently but contributes to the overall system function, allowing large-scale implementations to maintain both simplicity and performance.
2Productivity
If a top-down synthesis approach is used, then system objective function is achieved, but the inherent spiking dynamics of neurons are lost
Solution Approach 1:
The patent implements dynamic behavior in the mirrored neuron pairs through time-varying connections and adaptive responses. The neurons exhibit spiking dynamics and temporal patterns that emerge from the interaction between mirrored pairs, allowing the system to achieve objectivity while preserving biologically plausible spiking behavior rather than using static simplified models.
Solution Approach 2:
The mirrored neuron pairs are connected through feedback pathways that allow each neuron to adjust its response based on the activity of its mirror partner and the overall network state. This feedback mechanism enables the system to achieve system objectives while maintaining inherent spiking dynamics and self-optimizing behavior.
3Device complexity
If individual neuron responses are simplified to statistical properties, then device complexity is reduced, but the relationship between spike shape and system objective becomes unclear
Solution Approach 1:
The patent adds a temporal dimension to the neuron responses by using mirrored pairs that interact over time. Instead of simplifying responses to static statistical properties, the system captures temporal patterns and dynamic interactions between mirrored neurons. This dimensional expansion allows the system to preserve spike dynamics information while using simpler individual neuron models.
Solution Approach 2:
The patent creates a composite neural response by combining the outputs of mirrored neuron pairs. The system leverages the complementary properties of the paired neurons to achieve richer representations than individual neurons could provide alone, maintaining spike dynamics information through the composite behavior of the pairs.
Data Source
AI summary
A growth transform neural network system is disclosed that includes a computing device with at least one processor and a memory storing a plurality of modules, including a growth transform neural network module, a growth transform module, and a network convergence module. The growth transform neural network module defines a plurality of mirrored neuron pairs that include a plurality of first components and a plurality of second components. Each first and second component is connected by a normalization link. The first components are interconnected according to an interconnection matrix, and the second components are interconnected according to the interconnection matrix. The growth transform module updates each first component of each mirrored neuron pair according to a growth transform neuron model. The network convergence module converges the plurality of mirrored neuron pairs to a steady state condition by solving a system objective function subject to at least one normalization constraint.