Reproduce “Sequential Experimental Design for X-Ray CT Using Deep Reinforcement Learning” on OpenResearch
Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.
Universal app for Apple silicon and Intel. Requires macOS 11 or later.
About this paper
Tianyuan Wang, Felix Lucka, Tristan van Leeuwen
Researchers at CWI developed a deep reinforcement learning framework for sequential X-ray CT angle selection, aiming to reduce the number of necessary projections while maintaining reconstruction quality, especially for industrial in-line inspection. The approach consistently outperformed equidistant angle selection for objects with non-uniform informative angles, enabling adaptive and faster scanning.
Key insights
- Deep reinforcement learning provides a viable framework to learn non-greedy, adaptive sequential experimental design policies, effectively bypassing the computational challenges of classical OED methods for real-time CT.
- Representing the belief state with the current image reconstruction, rather than raw measurements, enables efficient feature extraction for industrial scenarios characterized by few angles and diverse object variations.
- The choice of reward function significantly impacts policy learning, with end-to-end rewards leading to higher average performance and incremental rewards facilitating faster training convergence.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.