SOHAM SARKAR

SOHAM SARKAR

he / him

Ph.D. candidate with a strong background in robotic control algorithms, deep reinforcement learning, convex optimization, advanced system integration, with hands-on experience in deploying controllers in robotic equipments.

Open-Sources

Tracking-Control GitHub Repo stars
A model-free, machine-learning framework to control a robotic manipulator using only partially observed states, where the controller is realized by reservoir computing. The effectiveness is demonstrated using a variety of periodic and chaotic signals.

Reservoir-Computing-and-Hyperparameter-Optimization GitHub Repo stars
Reservoir computing (echo state network) for short- and long-term prediction of chaotic systems, with tasks Lorenz and Mackey-Glass systems as examples. Bayesian optimization (hyperparameter optimization algorithm) is used to tune the hyperparameters and improve the performance.

Parameter-Tracking-with-Machine-Learning GitHub Repo stars
Tracking parameters within the system from which only partial state observation is available.

Dynamics-Reconstruction-ML GitHub Repo stars
Bridging known and unknown dynamics. Training on a diverse of dynamical systems, and testing on new unseen target systems with sparse and random observations.

AMOC GitHub Repo stars
Reservoir computing to predict the Atlantic Meridional Overturning Circulation (AMOC) evolution in short term.

Power-grid-attack-detection-and-state-estimation-with-machine-learning GitHub Repo stars
Detecting attacks and estimating states of power grids from partial observations with machine learning.

Meta-learning-Ecosystems GitHub Repo stars
Learning to learn ecosystems from limited data - a meta-learning approach

Dynamical-Systems-Control-with-Machine-Learning GitHub Repo stars
Controlling dynamical systems by reservoir computing, with chaotic Lorenz system as an example.