"Computers and tiny things" - A look into generative models and nano-tweezers

Image and text generating models took the internet by storm in recent years, but research in the area of generative models is not new. Beyond the well known examples, such models have also been used for anything from creating new molecules for drug development to testing hypotheses in astrophysics. However, many of those models take many steps to create viable outputs, with each step costing time and energy. In this talk I will illustrate what causes these inefficiencies, present how conditional flow matching models may reduce the number of steps needed, and describe my own research in improving them further.

Cells make tiny mechanical movements as they grow and function—just nanometers in size, and incredibly weak. These motions are hard to measure, especially in single cells. I use plasmonic optical tweezers, a laser-based tool, to detect these subtle movements in individual bacteria. This helps us better understand their behavior, growth, and interaction with their environment—including how they respond to antibiotics.