BEGIN:VCALENDAR
VERSION:2.0
PRODID:-//Memento EPFL//
BEGIN:VEVENT
SUMMARY:Structure Aware Lagrangian Relaxations with Approximation Guarante
 es for Weakly Coupled Markov Decision Processes
DTSTART:20240422T140000
DTEND:20240422T150000
DTSTAMP:20260611T110154Z
UID:2fd378a016cc6802bac3431c3da378861b66f0c329a9bd3f637447c0
CATEGORIES:Conferences - Seminars
DESCRIPTION:Professor Selva Nadarajah\n\nUniversity of Illinois Chicago (U
 IC)\nAbstract: Weakly coupled Markov decision processes represent a struct
 ured class of stochastic decision models that couple component Markov deci
 sion processes via linking constraints\, and arise in resource allocation\
 , project selection\, and scheduling problems. They are approximated using
  Lagrangian relaxation\, which averages combinatorial information in the l
 inking constraints\, as well as fluid models and approximate linear progra
 mming\, which capture this information exactly\, often with higher computa
 tional complexity. We introduce structure-aware Lagrangian relaxations\, a
  hierarchy of models that contains these two extremes studied in the liter
 ature\, and unexplored models in between. We construct this hierarchy by t
 aking a novel polyhedral perspective towards the representation of combina
 torial structure of linking constraints. We prove a transferability result
  that ports integrality ratio guarantees of combinatorial systems to appro
 ximation guarantees on the optimal policy upper bound of models in our hie
 rarchy with respect to the best model. For the broad class of packing prob
 lems\, we show that model size depends exponentially and polynomially on t
 he number of components when coupling constraints are represented exactly 
 and approximately\, respectively\, a striking difference that supports fin
 ding models of small size with good performance in our hierarchy. For reso
 urce allocation problems\, we numerically observe that structure-aware Lag
 rangian relaxations with bounds and policies comparable to existing models
  are orders of magnitude smaller in size\, allowing us to improve the stan
 dard Lagrangian for previously intractable instances. (Joint work with And
 re Cire\, University of Toronto\; talk based on paper at this link forthco
 ming in Management Science and follow up work)\n\n\n\n\nBio sketch: Selva 
 Nadarajah is an Associate Professor of Information and Decision Sciences a
 t the University of Illinois Chicago (UIC) College of Business and the Dec
 ision Intelligence R&D Lead at the Discovery Partners Institute (Innovatio
 n Hub of the University of Illinois System). He is also Guest Faculty at A
 rgonne National Laboratories. Selva develops dynamic valuation and decisio
 n-making models that guide investment\, operations\, and risk management\,
  with a focus on the energy industry. His ongoing energy projects study th
 e effect of limited market data on NetZero planning and the consideration 
 of social/equity risks. Selva also develops self-adapting stochastic optim
 ization and reinforcement learning methods that endogenize model reformula
 tion\, performance acceleration\, and parameter selection\, with the goal 
 of making these tools accessible to non-technical experts\, drawing inspir
 ation from commercial deterministic optimization solvers. Selva's research
  has been recognized with the 2021 Commodity and Energy Markets Associatio
 n (CEMA) Best Paper Award\, the 2020 INFORMS ENRE Young Researcher Prize\,
  the Best Overall Paper at the 2020 NeurIPS Workshop on Tackling Climate C
 hange with Machine Learning\, and the 2014 William L. Cooper Dissertation 
 Award.\n\n 
LOCATION:ODY 4 03 https://plan.epfl.ch/?room==ODY%204%2003 https://epfl.zo
 om.us/j/65607081423
STATUS:CONFIRMED
END:VEVENT
END:VCALENDAR