ÖGOR Summer-School for PhD-candidates and Post-Docs
The Austrian Society for Operations Research (OeGOR) organizes for the third time a summer-school for PhD-candidates and Post-Docs.
Invited Speakers: Timo Gschwind (RPTU Kaiserslautern):
Introduction to Column-Generation Theory and Application
Maximilian Schiffer (TU München):
Combinatorial optimization augmented machine learning for contextual multi-stage problems
Date: 22.07.2024 – 26.07.2024
Place: University for Continuing Education (Donau Uni) Krems
Participation in the school is free of charge, if your institution or supervisor are members of ÖGOR or of an OR-Society in a neighboring country. However, lunch will be organized centrally, thus requiring a meal contribution of ca. € 60.- (nonrefundable) to be paid until 21.06.2024. All other food, travel and hotel arrangements must be organized and covered by the participants.
Registration: send an e-mail until May 24, 2024 to raimund.kovacevic@donau-uni.ac.at
Each invited speaker will give a two-day program. The participants are also invited to give a presentation of their work. Take the chance to create a bond with fellow students and with OeGOR!
22.7.-23.7. Timo Gschwind:
Introduction to Column-Generation Theory and Application
Column Generation (CG)-based methods are (among) the leading exact solution techniques
for many important combinatorial optimization problems including problems in routing,
scheduling, cutting and packing, and network design to name a few. CG itself is an iterative
procedure that can tackle Linear Programs (LPs) containing a huge number of variables. At
each iteration, it solves a restricted version of the LP containing only a subset of the variables
and one or several pricing problems to generate missing variables. When solving integer
problems, CG is embedded in an overall strategy (e.g., Branch-and-Bound) to guarantee
integer solutions, resulting in an integer CG approach (e.g., Branch-and-Price, B&P). However,
(integer) CG approaches are not out-of-the-box solution methods. Many important building
blocks are often problem specific and a textbook implementation of basic CG is typically not
competitive. This course begins with a practice-oriented introduction to CG and B&P. It then covers the
basic theory of (integer) CG and discusses more advanced techniques such as pricing
strategies, cutting, strategies to obtain integer solutions, stabilization, variable fixing, or CGbased
heuristics. Overall, the course aims at providing a balance between theoretical foundations and ready-to-apply knowledge for designing and implementing CG approaches.
The course also includes practical implementation sessions.
24.7.: Short presentations by the participants
25.7.-26.7. Maximilian Schiffer:
Combinatorial optimization augmented machine learning for contextual multi-stage problems
Combinatorial optimization augmented machine learning (COAML) is a novel field that
combines methods from machine learning and operations research to tackle contextual datadriven
problems that involve both uncertainty and combinatorics. These problems arise
frequently in industrial processes, where firms seek to leverage large and noisy data sets to
optimize their operations. COAML typically involves embedding combinatorial optimization
layers into neural networks and training them with decision-aware learning techniques.
In this summer school, we will learn i) how such algorithmic pipelines work and ii) how they
can be designed depending on the planning problem faced. We will then iii) focus on the
difference between pipelines that are based on structured learning and pipelines that are
based on reinforcement learning. Finally, we will focus iv) on a hands-on coding exercise to
build a respective algorithmic pipeline on our own to reflect the theory studied.
The participants are invited to attend the course with their own laptops, having
super-user access on their computers for installing the required resources, and
with wifi-internet access (Eduroam or 4G/5G).
