Spike-based population coding and working memory.

Area of Science:

• Computational Neuroscience
• Neural Coding
• Decision Making

Background:

• Human behavior demonstrates optimal decision-making under uncertainty in perceptual and motor tasks.
• A fundamental question in neuroscience is how neural populations compute probabilities using spiking neurons.

Purpose of the Study:

• To develop a comprehensive framework for optimal, spike-based sensory integration and working memory in dynamic environments.
• To propose a model where probability distributions are inferred spike-per-spike by recurrently connected integrate-and-fire neurons.

Main Methods:

• Development of a theoretical framework for spike-based probabilistic inference in neural networks.
• Modeling recurrently connected integrate-and-fire neurons to simulate sensory integration and working memory.
• Analysis of spike train variability and comparison with rate-based coding models.

Main Results:

• Networks can optimally combine sensory cues, track time-varying stimuli, and maintain evidence over extended periods.
• Population activity and working memory states represent entire probability distributions, not single values.
• Spike times deterministically signal prediction errors, enabling robust encoding despite neural noise and variability.

Conclusions:

• Spike-based inference provides a mechanism for optimal probabilistic computation in neural systems.
• This coding scheme allows for robust working memory and sensory integration, outperforming traditional rate-based models.
• Variability in spike trains is a feature of optimal inference, not merely noise.