Eric Roberts Email & Phone Number

Schenectady, NY, US

Niskayuna, New York, United States

• I build end-to-end machine/deep learning frameworks and solutions for large-scale image analysis tasks in many collaborations across an array of biological and x-ray scattering sciences. In my total time at Lawrence.
• Lead developer of DLSIA, or Deep Learning for Scientific Image Analysis, a comprehensive Python-based library focused on building deep learning workflows with user-defined architectures for scientific image analysis.
• In addition to coding development, I built and currently maintain all Git and ReadTheDocs web development (via Git/Sphinx sychronization) for showcasing and disseminating DLSIA tuto-rials and use-cases.
• Employed sparsity techniques to augment existing CNN architectures, resulting in up to a 90% reduction in model size and inference speed from standard U-Net architectures.
• Multi sparse network ensembling for semantically segmenting popular STAREdataset resulted in.83 F1 evaluation score, within 1.1% of current state-of-the-art,
• Project lead in expanding automated label curation pipelines and applying super-vised deep learning techniques for sparse-label, pixel-by-pixel segmentation of sub-cellular structures on the sub-nanometer scale.

1 Cyclotron Road, Berkeley, CA

Within Molecular Biophysics & Integrated Bioimaging (MBIB) and Applied Math (CAMERA) divisions.Co-Developer of MLExchange, a DOE-funded shared MLOps platform for image analysis tasks, on-the-fly label curation, and pre-trained model distillation (i.e. transfer learning).Investigated numerical integration techniques for sparse-sampled quadratures of maximum.

Within Advanced Light Source (ALS) division.Augmenting existing CNN architectures, achieved a 99.1% sorting rate for metadata retrieval of x-ray scattering images across several modalities. Metadata preservation and database characterization using MongoDB.

Merced, California

My research fields included nonlinear dynamical systems, topological fluid dynamics, and mixing in viscous fluids. More specifically, by re-framing existing techniques in a computational geometric lens, I developed and implemented algorithms for extracting meaningful topological information and quantifying chaotic mixing in highly dynamic and nonlinear.

Responsibilities include the design and implementation of a three-week 2017 Summer course in numerical analysis for second-year undergraduate students and a three-week 2018 Summer course serving as an introduction to data science and machine learning for third-year undergraduate students. Students from both years have taken inspiration from our summer work.

Merced, California

University Of Minnesota

Six week session designed to provide graduate students in Mathematics with training that is valuable outside of academia. Daily technical skill building modules on machine learning techniques, statistical modeling, and optimization methods for decision-making processes culminated in two group capstone projects involving open-ended problems posed by.

Yotba Linda, California

Doctor Of Philosophy (Ph.D.), Applied Mathematics

Bachelor Of Science (B.S.), Applied Mathematics

Applied Mathematics