BEGIN:VCALENDAR PRODID:-//eluceo/ical//2.0/EN VERSION:2.0 CALSCALE:GREGORIAN BEGIN:VEVENT UID:b241be2eb6010acce6dff627e88840b2 DTSTAMP:20260521T222040Z SUMMARY:Deep Learning Models with Hard Physical and Logical Constraints DESCRIPTION:\nTitle: Deep Learning Models with Hard Physical and Logical Co nstraints\nAbstract: Modern deep learning models have achieved remarkable success in computer vision\, natural language processing\, and a growing r ange of scientific applications\, from image analysis to molecular and mat erials modeling.  However\, their direct application to chemical engineer ing problems remains challenging because engineering systems are governed by strict physical and logical constraints and are often characterized by sparse\, expensive data.\nStandard neural networks may produce accurate pr edictions on average\, yet still violate fundamental principles such as ma ss and energy balances\, operational limits\, or logical rules\, making th em unreliable for scientific and industrial use. In this talk\, I will pre sent a framework for embedding hard physical and logic constraints directl y into deep learning models using ideas from constrained optimization. Ins tead of enforcing constraints through penalty terms or post-processing\, w e introduce differentiable projection layers inspired by KKT conditions an d convex optimization that guarantee constraint satisfaction by constructi on.\nThese layers are model-agnostic\, computationally efficient\, and com patible with standard training via backpropagation. I will illustrate the approach through chemical engineering case studies\, including surrogate m odeling of chemical processes with exact mass balance enforcement\, as wel l as broader applications involving inequality and logic constraints. Resu lts show improved data efficiency\, stronger generalization\, and inviolab le feasibility compared to unconstrained and softly constrained models. Ov erall\, this work highlights how optimization-inspired neural architecture s can bridge the gap between data-driven learning and first-principles mod eling\, enabling reliable machine learning tools for safety-critical chemi cal engineering applications.\nBio: Can obtained his bachelor’s degree f rom Tsinghua University\, China\, in Chemical Engineering. He completed hi s PhD in Chemical Engineering at Carnegie Mellon University. His PhD resea rch is focused on stochastic mixed-integer nonlinear programming and long- term expansion planning of power systems. Can did a one-year Postdoc at Po lytechnique Montreal on using machine learning techniques to accelerate op timization algorithms. He joined the Davidson School of Chemical Engineeri ng at Purdue University as an assistant professor in the Fall of 2022. His research group is focused on optimization\, machine learning\, and applic ations in sustainable energy systems. His group won Air Liquide’s global scientific challenge on data sharing for decarbonization in 2023\, the Am azon Research Award in 2024\, and the NSF CAREER Award in 2025. URL:https://www.imperial.ac.uk/events/209930/deep-learning-models-with-hard -physical-and-logical-constraints/ DTSTART;TZID=Europe/London:20260608T160000 DTEND;TZID=Europe/London:20260608T170000 LOCATION:LT2\, ACE Extension\, South Kensington Campus\, Imperial College L ondon\, London\, SW7 2AZ\, United Kingdom END:VEVENT BEGIN:VTIMEZONE TZID:Europe/London BEGIN:DAYLIGHT DTSTART:20260608T160000 TZNAME:BST TZOFFSETTO:+0100 TZOFFSETFROM:+0100 END:DAYLIGHT END:VTIMEZONE END:VCALENDAR