Sreejeet Maity
@sreejeetm1729ποΈGraduate Research Assistant | Ph.D. Candidate @ North Carolina State University | Interested in Optimization, Reinforcement Learning, and Probability.
Language Breakdown
Lines of code distribution across 15 owned repositories
I-Shaped Developer
I-shapedSpecialist β deep expertise in Jupyter Notebook
Collaboration Network
Global Impact visualization
Repos
28
PRs
0
Growth
+18%
Top Collaborators
No collaborator data yet.
Coding Streak
Contribution activity over the past year
Xinyi Liu
@liuxy10
Haimin Hu
@HaiminClack
ameyanjarlekar
@ameyanjarlekar
Manan Tayal
@tayalmanan28
Omkar Patil
@omkarpatil18
Top Repositories
Reinforcement learning (RL) has shown promise in financial applications such as algorithmic trading. However, the high-dimensional nature of market data introduces the curse of dimensionality, making neural network-based function approximators essential. Contributors: Sreejeet Maity, Sai Kavya Marthala, Rajesh Debnath.
Accepted at ICML 2026π. ππππππ π°ππ’ππ-π/π°ππ’ππ-ππ°π/πΌ: The first provably robust variants of asynchronous Q-learning that tolerates adversarially corrupted rewards. Our algorithm is distribution-agnostic, and achieves near-optimal finite-time guarantees up to a provably unavoidable corruptive term.
Accepted at ACC 2026 π. π ππ³π : We propose and analyze a new algorithm that achieves collaborative speedups in sample complexity for Q-learning over static and time-varying networks.
Accepted at IEEE CDC 24 π. We analyze Q-learning's robustness against strong-contamination attacks that disrupt reward signals. Our robust Q-learning algorithm employs historical data to create resilient Bellman operators, achieving finite-time convergence rates that match existing bounds, with a minor error linked to the corruption fraction.
Config files for my GitHub profile.
Research Website
π΅ππππππππ πΌπ°ππ»-πΆπ’π : We introduce a custom multi-agent reinforcement learning environment built with Gymnasium and Pygame, designed for evaluating federated RL (FRL) algorithms. The environment models a grid world where multiple agentsβsuch as robotsβnavigate to accomplish spatially distributed tasks, like reaching delivery points.
Accepted at ACC 26 π. ππππππ π΅ππ-π : A federated Q-learning algorithm that stays reliable even when a fraction of agents are adversarial. It blends model-based/free updates to ensure (i) high-probability exact convergence in the infinite-sample limit and (ii) near-optimal finite-time rates with collaborative sample-complexity.
A collection of economics and finance papers that adopt reinforcement learning as a solution method.
Implementation of all RL algorithms in a simpler way
Open Source Impact
Contributions to external projects
No external contributions found.