| Abstract: | Ptychography provides high-contrast quantitative imaging without prior object information, but conventional approaches face limitations in acquisition time and resolution. I will present novel computational imaging methodologies overcoming these constraints through physics-informed deep learning. First, “deepSSP” achieves superior single-shot ptychography reconstruction using only experimental data, demonstrating 1.25× resolution enhancement beyond theoretical limits with millisecond processing. Second, “deepTIMP” enables ultrafast imaging of dynamic phenomena through physics-informed neural networks, successfully reconstructing multiple temporal frames with exceptional robustness. Third, “SAIDAST” addresses high-resolution space imaging through distributed telescope arrays, achieving 2.38× resolution enhancement with O(1) memory complexity and superior noise resistance. These advances establish a new paradigm where physics-informed deep learning fundamentally extends ptychographic capabilities, enabling applications from ultrafast microscopy to space telescopes while providing interpretability and performance advantages over traditional approaches. |