BEGIN:VCALENDAR
VERSION:2.0
PRODID:-//sebbo.net//ical-generator//EN
URL:https://lectures.london/imperial-college/deep-learning-models-with-har
 d-physical-and-logical-constraints/calender.ics
NAME:Lectures London
X-WR-CALNAME:Lectures London
TIMEZONE-ID:Europe/London
X-WR-TIMEZONE:Europe/London
BEGIN:VEVENT
UID:6a9d04f7-f69e-45af-bab2-bc7ee2d76abc
SEQUENCE:0
DTSTAMP:20260604T073406
DTSTART:20260608T150000
DTEND:20260608T160000
SUMMARY:Deep Learning Models with Hard Physical and Logical Constraints
LOCATION:Imperial College: LT2\, ACE Extension
DESCRIPTION:Title: Deep Learning Models with Hard Physical and Logical Con
 straints\nAbstract: Modern deep learning models have achieved remarkable s
 uccess in computer vision\, natural language processing\, and a growing ra
 nge of scientific applications\, from image analysis to molecular and mate
 rials modeling.  However\, their direct application to chemical engineeri
 ng problems remains challenging because engineering systems are governed b
 y strict physical and logical constraints and are often characterized by s
 parse\, expensive data.\nStandard neural networks may produce accurate pre
 dictions on average\, yet still violate fundamental principles such as mas
 s and energy balances\, operational limits\, or logical rules\, making the
 m unreliable for scientific and industrial use. In this talk\, I will pres
 ent a framework for embedding hard physical and logic constraints directly
  into deep learning models using ideas from constrained optimization. Inst
 ead of enforcing constraints through penalty terms or post-processing\, we
  introduce differentiable projection layers inspired by KKT conditions and
  convex optimization that guarantee constraint satisfaction by constructio
 n.\nThese layers are model-agnostic\, computationally efficient\, and comp
 atible with standard training via backpropagation. I will illustrate the a
 pproach through chemical engineering case studies\, including surrogate mo
 deling of chemical processes with exact mass balance enforcement\, as well
  as broader applications involving inequality and logic constraints. Resul
 ts show improved data efficiency\, stronger generalization\, and inviolabl
 e feasibility compared to unconstrained and softly constrained models. Ove
 rall\, this work highlights how optimization-inspired neural architectures
  can bridge the gap between data-driven learning and first-principles mode
 ling\, enabling reliable machine learning tools for safety-critical chemic
 al engineering applications.\nBio: Can obtained his bachelor’s degree fr
 om Tsinghua University\, China\, in Chemical Engineering. He completed his
  PhD in Chemical Engineering at Carnegie Mellon University. His PhD resear
 ch is focused on stochastic mixed-integer nonlinear programming and long-t
 erm expansion planning of power systems. Can did a one-year Postdoc at Pol
 ytechnique Montreal on using machine learning techniques to accelerate opt
 imization algorithms. He joined the Davidson School of Chemical Engineerin
 g at Purdue University as an assistant professor in the Fall of 2022. His 
 research group is focused on optimization\, machine learning\, and applica
 tions in sustainable energy systems. His group won Air Liquide’s global 
 scientific challenge on data sharing for decarbonization in 2023\, the Ama
 zon Research Award in 2024\, and the NSF CAREER Award in 2025.
URL;VALUE=URI:https://www.imperial.ac.uk/events/209930/deep-learning-model
 s-with-hard-physical-and-logical-constraints/
END:VEVENT
END:VCALENDAR