off-policy

This repository contains the implementation of a wide variety of Reinforcement Learning Projects in different applications of Bandit Algorithms, MDPs, Distributed RL and Deep RL. These projects include university projects and projects implemented due to interest in Reinforcement Learning.

PyTorch implementation of "Sample-efficient Imitation Learning via Generative Adversarial Nets"

My content of CS294 Deep Reinforcement Learning course, conduced by Sergey Levine from UC Berkeley.

PyTorch implementation of our work: "Lipschitzness Is All You Need To Tame Off-policy Generative Adversarial Imitation Learning"

AI project combining Monte Carlo racetrack control with on/off-policy learning, weighted importance sampling, and Sinkhorn optimal-transport face morphing with Wasserstein interpolation.

Q-learning is an off-policy temporal-difference control algorithm. It learns the value of the optimal action, independent of the action actually taken by the agent.

Ensemble and Auxiliary Tasks for Data-Efficient Deep Reinforcement Learning

Off-policy Monte Carlo control (weighted importance sampling) solving Sutton & Barto's Racetrack problem — with a training CLI, wandb logging, weight saving, and side-by-side agent replays.

This repository contains all of the Reinforcement Learning-related projects I've worked on. The projects are part of the graduate course at the University of Tehran.

Sample Policy Gradient

🧗Comparative Reinforcement Learning analysis implementing SARSA (On-Policy) vs Q-Learning (Off-Policy) from scratch on OpenAI Gymnasium CliffWalking-v1 with live side-by-side animations and policy heatmaps.

Contains PyTorch Implementation of the following off policy actor critic algorithms

Autonomous Parking with Deep Reinforcement Learning Custom MDP Development

Retroactive Advantage Correction (RAC): closed-form V-trace bias correction for delay-aware RLHF — RLxF Workshop @ ICML 2026

DeepMind ACER 논문 기반 Actor-Critic + Experience Replay를 PyTorch로 구현하고, CartPole-v1에서 Retrace·Truncated IS·Bias Correction·Trust Region을 적용해 학습합니다.

Fine-Tuning GPT-2, RoBERTa, and PPO Architecture for Enhancing Reliability of General-Purpose Chatbot

Collection of codes pertaining to my research in model-free RL algorithms.

PyTorch implementation of our work: "Where is the Grass Greener? Revisiting Generalized Policy Iteration for Offline Reinforcement Learning"

PROJECT MIGRATED TO CODEBERG - Reinforcement Learning in Multiplicative Domains

PyTorch implementation of our work: "Optimality Inductive Biases and Agnostic Guidelines for Offline Reinforcement Learning"