Computational Imaging Automation
The emergence of advanced computational imaging techniques has revolutionized microscopy, enabling unprecedented insights into material structures and chemical compositions at the nanoscale. As scientific facilities like the Advanced Photon Source undergo major upgrades to achieve orders of magnitude higher data collection rates, we face a fundamental challenge: the traditional paradigm of human-centric microscopy workflows is becoming unsustainable and can limit the full potential of these technological breakthroughs.
Our research aims to revolutionize traditional microscopy practices and transform them into automated and intelligent smart microscopy. We envision in the future artificial intelligence agents work alongside scientists, not just executing predetermined protocols, but actively participating in the experiment process—from experiment condition optimization to data analysis to result interpretation.
The cornerstone of our approach is the seamless integration of physics-based modeling with artificial intelligence. Rather than treating microscopy automation as a pure machine learning problem, we recognize that the underlying physics provides crucial constraints and insights that can guide automation strategies. This Physics-Informed AI paradigm enables us to develop automation solutions that are both data-efficient and scientifically rigorous.
Our current research focuses on advancing this vision through several innovative directions: 1. Developing differentiable physics models that enable end-to-end optimization of complex analysis workflows 2. Pioneering physics-informed Bayesian optimization approaches for automated experimental design 3. Creating multi-agent AI systems that can reason about microscopy experiments much like human experts do
Our ultimate goal is to establish a new paradigm for scientific imaging where human creativity and expertise are amplified by AI-driven automation. By removing human bottlenecks while maintaining scientific rigor, we enable higher throughput experimentation, more consistent analysis, and the potential for new scientific discoveries that might be missed in traditional manual workflows.
Back to top