A Survey of Methods for Automated Algorithm Configuration 

   Tutorial at the
International Conference on Automated Machine Learning (AutoML) 


In many fields of industry or academia computationally hard problems, such as constraint satisfaction problems, Boolean satisfiability problems (SAT) or vehicle routing problems must be solved regularly. The solution quality of algorithms that solve these problems and even their required runtime is often highly influenced by their internal parameters. Thus, it is crucial to adjust these parameters and find a parameter configuration that leads to optimal performance of the so-called target algorithm. 

However, optimizing them by hand is a complex or even infeasible task for the user, which emerged in the research field of Automated Algorithm Configuration (AC). Broadly speaking, approaches in AC aim to find an optimal parameter configuration for a given target algorithm concerning a given target metric such as runtime or solution quality that should be optimized. In the last decades, many algorithms and problem variants have been proposed in this field, which come with different characteristics and challenges. 

This tutorial aims to provide an overview of different variants of both AC problems and methods, i.e., algorithm configurators, to tackle AC problems. To this end, we discuss two classification schemes: one for AC problems, and one for algorithm configurators. Based on these, we structure and summarize the available literature and classify existing problem variants as well as approaches to AC.

PRESENTERS

Viktor bengs

LMU Munich, Germany

Jasmin Brandt

Paderborn University, Germany

MArcel WEver

Munich Center for Machine Learning,
LMU Munich, Germany

SCHEDULE

This tutorial will be held in-person at the AutoML Conference at the Hasso Plattner Institute, Potsdam, Germany on September 13th 2023.

In our tutorial, we give an overview of different variants of both AC problems and methods, i.e., algorithm configurators, to tackle AC problems. To this end, we discuss two classification schemes: one for AC problems, and one for algorithm configurators. Based on these, we structure and summarize the available literature and classify existing problem variants as well as approaches to AC.

The program schedule will be as follows.

OUTLINE

TARGET AUDIENCE

With the tutorial we aim to attract both researchers that are already active in one of the above domains, as well as researchers with little or no prior experience in algorithm configuration. We will only assume a general familiarity with well-known machine learning techniques for classification and regression (neural networks, probabilistic models, risk minimization, etc.). Therefore, the tutorial is of interest to scholars from diverse subfields of machine learning and with different backgrounds. In particular, anyone who is interested in an overview of AC, its problem variants and the existing methods that can solve these have come to the right place.


PRACTICAL INFORMATION

For registration, see the AutoML website.


SLIDES

Slides are available here: Link


VENUE (September 13th, Room TBA)

Hasso Plattner Institute, Potsdam, Germany

Map View