A research project exploring deep learning as a way to build an automatic, experiment-specific framework for physical discovery.
It represents an earlier research direction within the RSC program.
A reinforcement learning study that uses the Rubik's Cube as a hard-exploration environment. The work compares standard PPO, manually designed curriculum learning, and an online teacher approach based on an epsilon-decay-greedy strategy.
A study exploring explainability techniques for medical imaging, including Gradients, Integrated Gradients, XRAI, Grad-CAM, and Counterfactual Explanations. The work also examines sanity checks and execution-cost considerations.
A computer vision project focused on detecting road lanes and vehicles for autonomous driving applications.
