The Role of E/I modulation in Shaping Neural Dynamics in Decision-Making and Adaptive Learning

Overview

This repository contains an analysis of the role of excitation and inhibition in shaping neural dynamics in two settings:

• Decision-Making* --- using a mean-field Wilson-Cowan rate model
• Adaptive Learning in Dynamic Environments (Continual Reinforcement Learning in a changing sequence of Probabilistic Reversal Learning (PRL) tasks) --- using a Neural Actor Critic model with an E/I modulation mechanism

Directory Structure

1. mean_field__decision_making/

This folder contains notebooks and scripts for simulating and analyzing neural dynamics in decision-making processes using mean-field approximations.

Files:

• evidence_accum_perf_crit.ipynb: For understanding evidence accumulation in achieving performance criteria during decision-making tasks.
• inhib.ipynb: Investigates the impact of inhibition on neural dynamics.
• plotting.py: Helper functions for visualizing results and simulations.
• recur_excit.ipynb: Studies the role of recurrent excitation in decision-making dynamics.
• role_inhib.ipynb: Focuses on the role of inhibition in decision-making processes.
• sim_dyn.py: Core simulation script for dynamic systems modeling.
• utils.py: Utility functions for preprocessing, computation, and support tasks.

2. neuralAC__adaptive_learning/

This folder contains code for implementing and analyzing a Neural Actor-Critic (NeuralAC) framework for adaptive learning.

Files:

• baselines.ipynb: Compares the performance of over wE.
• environment.py: Defines the environment setup for reinforcement learning tasks.
• experiment.py: Contains code for running experiments and simulations.
• neural_actor_critic.py: Implements the Neural Actor-Critic algorithm.
• plotting.py: Visualization scripts specific to adaptive learning experiments.
• role_inhib_strong_recur.ipynb: Examines the role of inhibition in recurrent networks with high precision.
• role_inhib_weak_recur.ipynb: Examines the role of inhibition in recurrent networks with low precision.
• utils.py: Common utility functions used across experiments.

Getting Started

Prerequisites

• Python 3.8 or higher
• Recommended packages:
  • numpy
  • matplotlib
  • scipy
  • jupyter
  • pandas

Installation

1. Clone the repository:

   git clone <repository_url>
   cd EL_dynamics

2. Install dependencies:

   pip install -r requirements.txt

Running the Code

1. Navigate to the desired folder:

   cd mean_field__decision_making
   # or
   cd neuralAC__adaptive_learning

2. Open the notebooks in Jupyter:

   jupyter notebook <notebook_name>.ipynb

Key Features

• Simulations of mean-field neural dynamics for decision-making.
• Analysis of recurrent excitation and inhibition mechanisms.
• Neural Actor-Critic implementation for adaptive learning.
• Visualization tools for understanding results.