Gated spiking neural network using Iterative Free-Energy Optimization and rank-order coding for structure learning in...

Area of Science:

• Computational Neuroscience
• Cognitive Neuroscience

Background:

• The fronto-striatal system (PFC-BG) is crucial for executive functions, including working memory and sequence processing.
• Existing models often struggle to capture the dynamic and abstract nature of sequence representation in the brain.

Purpose of the Study:

• To develop a computational framework modeling the PFC-BG loop for audio memory sequence generation and recall.
• To propose and validate a novel coding strategy using gain modulation for abstract sequence representation.

Main Methods:

• Iterative free-energy optimization with spiking neural networks.
• Implementation of a gain-modulation mechanism for representing item rank and location in sequences.
• Development of the Inferno Gate model, an extension of the Inferno architecture.

Main Results:

• The model successfully encodes and retrieves audio memory sequences up to fifty items long.
• Gain modulation enables robustness to variability and long-term dependencies, mimicking gated recurrent neural networks.
• The proposed mechanism allows for the representation of novel sequences based on learned temporal structures.

Conclusions:

• Inferno Gate provides a biologically plausible model for working memory in the PFC-BG loop.
• The gain-modulation strategy offers a novel approach to abstract sequence representation in neural networks.
• The framework has potential applications in structural learning, goal-directed behavior, and hierarchical reinforcement learning.