Click for Demo

• Awarded Best Paper for research thesis titled Damped Sinusoidal Exploration Decay Schedule to improve Deep Q-Networks-based agent performance.
• As sole student author, proposed a novel damped sinusoidal equation to perform exploration decay instead of linear epsilon decay by a constant factor to optimize DQN-based reinforcement learning agents in sparse-reward environments. Utilized the Google Colab virtual machine powered by a NVIDIA Tesla K80 GPU and 13 GB RAM to process terabytes of data over 3500 episodes in over 16 training cycles each lasting up to 48 hrs.
• Presented at the 4th International Conference on Artificial Intelligence and Evolutionary Computations in Engineering Systems 2019 organized in association with Springer, IEEE and IET from 11th to 13th April, chaired by Dr. Ramazan Bayindir.
• Accepted for publication in Springer Advances in Intelligent Systems and Computing (AISC) series.

The code for the results of paper titled Damped Sinusoidal Exploration Decay Schedule to improve Deep Q-Networks-based agent performance is available in the research folder.

The Atari and Classic Control folders contain the project code before the research component.