Undergraduate Research — The Data Mine @Purdue

Worked on a research project exploring the intersection of quantum computing and reinforcement learning, focusing on how quantum methods compare to classical approaches in sequential decision-making problems.

Implemented both a classical Deep Q-Learning agent (PyTorch) and a quantum reinforcement learning model using variational quantum circuits in PennyLane, evaluating performance on the CartPole environment. The study analyzed training dynamics, convergence behavior, and overall efficiency using metrics such as episode duration and area under the curve (AUC).

Found that while both models achieved similar final performance, the quantum model exhibited distinct training characteristics—particularly sensitivity to initialization and the need for structured pre-training. Introduced a warm-up experience buffer to stabilize learning, significantly improving results.

This work highlighted the emerging potential of quantum reinforcement learning while also emphasizing current limitations due to simulation on classical hardware and the importance of algorithm design in quantum systems.

Fall 2025