Related Experiment Videos
Adaptive stochastic resonance in noisy neurons based on mutual information
1Department of Electrical Engineering, Faculty of Engineering, Thammasat University, Rangsit Campus, Klong Luang, Pathumthani 12120, Thailand. msanya@engr.tu.ac.th
IEEE Transactions on Neural Networks
|November 30, 2004
Summary
Noise can enhance signal processing in neurons through stochastic resonance (SR). This study demonstrates SR in various neurons and introduces a learning law to optimize noise levels for improved information throughput.
Area of Science:
- Computational Neuroscience
- Signal Processing
- Information Theory
Background:
- Noise can paradoxically enhance signal processing in certain neural systems.
- Stochastic resonance (SR) describes this phenomenon, where optimal noise levels improve information transfer.
- Understanding SR is crucial for developing more efficient artificial neural networks.
Purpose of the Study:
- To provide theoretical and simulation evidence for the stochastic resonance effect in individual noisy neurons.
- To introduce a novel, statistically robust learning law for identifying entropy-optimal noise levels.
- To investigate the robustness of adaptive SR against impulsive noise and varying input amplitudes.
Main Methods:
- Theoretical analysis and computer simulations of noisy threshold and continuous neurons.
- Utilizing histograms to estimate probability density functions during learning iterations.
- Developing and applying a new learning law to adaptively find optimal noise levels.
Main Results:
- Demonstrated that lone noisy neurons exhibit the SR effect, maximizing mutual information.
- Showcased a statistically robust learning law capable of finding entropy-optimal noise levels.
- Proved the robustness of adaptive SR against impulsive noise with infinite variance.
- Observed nonlinear behavior of optimal noise levels with increasing input signal amplitude.
Conclusions:
- Stochastic resonance is a viable mechanism for enhancing information processing in various types of neurons.
- The developed learning law effectively optimizes noise for improved neural signal throughput.
- Adaptive SR demonstrates resilience even under challenging, impulsive noise conditions.