Related Experiment Videos
Adaptive neural coding dependent on the time-varying statistics of the somatic input current
1Computation and Neural Systems Program, California Institute of Technology, Pasadena, CA 91125, USA.
Neural Computation
|December 1, 1999
Summary
This study presents a biologically plausible adaptive algorithm allowing neurons to self-optimize their firing dynamics. This method adjusts firing properties to match input current statistics, demonstrated in a silicon neuron.
Area of Science:
- Computational neuroscience
- Artificial neural networks
- Biophysics
Background:
- Nerve cells are assumed to optimize performance based on input statistics.
- Electronic neuron analogs require self-optimization for stable operation.
Purpose of the Study:
- To describe and demonstrate a biologically plausible adaptive algorithm for neuron self-optimization.
- To enable neurons to adapt firing dynamics to match input current characteristics.
Main Methods:
- Developed an adaptive algorithm for neuron self-optimization.
- Algorithm estimates somatic current from spike train via intracellular calcium concentration.
- Tested the algorithm in an analog VLSI-designed silicon neuron.
Main Results:
- The algorithm successfully adapts the neuron's current threshold and current-frequency relationship slope.
- Adaptation matches the mean (dc offset) and variance (dynamic range) of the somatic input current.
- Demonstrated continuous adjustment of neuronal firing dynamics.
Conclusions:
- The described adaptive algorithm enables neurons to self-optimize their firing dynamics.
- This principle of adaptation is biologically plausible and functional in silicon neurons.
- The algorithm allows neurons to match their firing properties to input current statistics.