Non linear systems in ML & Resorvoir Computing

This project, explores reservoir computing with a special focus on analyzing and using various non linear systems and hopefully able to create a general resorvoir for various different tasks.

Final Report

Seminar Presentation

TODO: Code Cleanup

Overview

Reservoir computing is a framework for designing recurrent neural networks, particularly effective for time-series processing and complex dynamic systems. This study investigates:

• The principles behind Echo State Networks (ESN)
• Advanced learning strategies and online/offline training methods
• Exploration of hyper-parameters optimization

One of the best resource for overview on RC: lit

Papers & Projects

Some literature:

• Many other in the papers directory
• Evolving Reservoirs for Meta Reinforcement Learning. (EvoAPPS 2024)
  HAL • Code
• From Implicit Learning to Explicit Representations.
  arXiv • PDF

Expt results

Below are some preliminary results of implementations and runs

From resorvoir_py.ipynb:

From simple_pendulum_RC.ipynb:

Using Logistic Map as a Reservoir

From using_lstm.ipynb

Acknowledgment

This project is conducted under the guidance of Prof. Gaurav Dar as part of my Study Oriented Project.