Modern microscopes create a data deluge with gigabytes of data generated each second, and terabytes per day. Storing and processing this data is a severe bottleneck, not fully alleviated by data compression. We argue that this is because images are processed as grids of pixels. To address this, we developed a content-adaptive representation of fluorescence microscopy images, the Adaptive Particle Representation (APR). The APR replaces pixels with particles positioned according to image content. The APR not only overcomes storage bottlenecks, as data compression does, but additionally overcomes memory and processing bottlenecks since the adaptivity can be used during processing tasks. In this talk, I will introduce the ideas and concepts of the APR, its performance on experimental data, and show how the APR can be used to enhance, rather than replace, existing algorithms and approaches, including applications to machine learning. Beyond image-processing I will also present how the APR can be used for adaptive resolution simulations, and discuss work on robust methods for data-driven model discovery for spatial-temporal data.
Speaker(s)
Bevan Cheeseman
Dr Bevan Cheeseman is an Applied Mathematician and recent PhD graduate from the Mosaic Group at the Max Planck Institute of Molecular Cell Biology and Genetics in Dresden, Germany. He is interested in using fluorescence imaging and a data-driven approach to understand how cells during development get to the right place, at the right time, wearing the right hat despite all the distractions. Such a data-driven approach requires dealing with the bottleneck of the massive processing and storage resources needed when using 3D+t images. To address this bottleneck, Bevan has developed a content-adaptive image representation, the Adaptive Particle Representation (APR). The APR uses spatially adaptive sampling focusing storage, memory, and computation resources to the important parts of the data in both space and time. By reducing required resources and processing time by factors of 10-100 the APR can enable high-throughput experimental and processing pipelines, which can form the basis of a data-driven approach. Bevan is also interested in the development of content-adaptive processing algorithms and data-driven model-discovery techniques that allow the discovery of driving dynamics (e.g. partial-differential equations) directly from experimental data.