Spike-Based Winner-Take-All Computation: Fundamental Limits and Order-Optimal Circuits

Area of Science:

• Computational Neuroscience
• Information Theory
• Neural Networks

Background:

• Winner-take-all (WTA) is a fundamental neural computation for selecting a small group of neurons from a larger pool.
• The robustness of spike-based WTA networks to input spike train randomness is not well understood.
• Understanding WTA network performance is key to deciphering brain computation.

Purpose of the Study:

• To analytically characterize the minimum waiting time for spike-based WTA networks to achieve a target decision accuracy.
• To investigate the impact of input spike train randomness on WTA network performance.
• To establish theoretical bounds and design efficient WTA circuits for neural computation.

Main Methods:

• Modeling spike-based WTA networks using independent Bernoulli processes for input spike trains.
• Deriving an information-theoretic lower bound on the waiting time required for a specific decision error.
• Designing a simple WTA circuit and analyzing its waiting time.

Main Results:

• An information-theoretic lower bound on waiting time was derived: [Formula: see text], dependent on rate sets and task difficulty.
• The derived lower bound is independent of the number of input trains, winners, and time slots.
• A simple WTA circuit achieved an order-optimal decision time of [Formula: see text], matching the lower bound's scaling.

Conclusions:

• The study provides a theoretical framework for understanding the time requirements of spike-based WTA networks.
• The designed WTA circuit demonstrates efficient and near-optimal performance in selecting neurons under noisy conditions.
• These findings contribute to the understanding of neural computation robustness and efficiency.