Independent researcher specializing in Reinforcement Learning, Deep RL, Machine Learning. Current Research Interests: Scalable and Efficient Exploration in Reinforcement Learning, and AI safety.Describe what you have done, what you are doing, and the kinds of things you are interested in.
• [Publication] Count-Based Exploration in Feature Space for Reinforcement Learning
Twenty-sixth International Joint Conference on Artificial Intelligence (IJCAI-17) https://www.ijcai.org/proceedings/2017/0344
• State Aggregation using Autoencoders in Reinforcement Learning
Tabular RL algorithms quickly looses viability when the state space is very large. In this project we aim to solve this problem by investigating the use of an auto-encoder neural network to learn relevant features thereby being able to compress the state space. https://github.com/surajx/Autoencoder-QLearner
• Proof Assistant: An online tool for writing Lemmon Style Natural Deduction proofs with real-time line-by-line proof checking. https://proof.surajx.in
• Modelling High-Dimensional Classification Problems Using Deep Learning: A study about the use of deep networks for feature extraction and classification on the 10,000 dimensional ARCENE dataset. https://github.com/surajx/autoencoders_arceneDescribe the most impressive thing you've done.