Tutorials
The role of the tutorials is to provide a platform for a more intensive scientific exchange amongst researchers interested in a particular topic and as a meeting point for the community. Tutorials complement the depth-oriented technical sessions by providing participants with broad overviews of emerging fields. A tutorial can be scheduled for 1.5 or 3 hours.
Tutorial proposals are accepted until:
October 4, 2023
If you wish to propose a new Tutorial please kindly fill out and submit this
Expression of Interest form.
![]()
Tutorial on
Combining Metaheuristics and Machine Learning to Design Fast and Reliable Algorithms
Instructor
|
Mario Pavone
University of Catania
Italy
|
|
|
|
Brief Bio
Mario Pavone is assistant professor in Computational Intelligence at the Department of Mathematics and Computer Science of University of Catania. His research is focused on the design and develops of nature-inspired algorithms, such as immunological algorithms and swarm intelligence heuristics, for optimization tasks in combinatorial, and systems biology problems. In particular, his main works have been focused on protein structure prediction in HP model; parameters extraction in biological networks (Gene Regulatory Network in the S-system model); graph coloring problem; and functions optimization. He is founder of the TaoScience Research Center and currently he is Chair of the Task Force on Artificial Immune Systems for the IEEE Computational Intelligence Society. Since February 2015 he is also Vice-Chair of the Task Force on Interdisciplinary Emergent Technologies for the IEEE Computational Intelligence Society.
|
Abstract
Abstract
The integration between machine learning and metaheuristics methods offers many advantages in all those problems where the use of purely stochastic algorithms is no longer sufficient. The literature on the hybridization of metaheuristics and machine learning can be broadly categorized into two groups: research papers where machine learning is used to enhance metaheuristics, and those where metaheuristics are used to improve the performance of machine learning techniques.
This tutorial presents applications where machine learning techniques are incorporated into both a Genetic Algorithm and an Immune-Inspired Algorithm to improve their performance. Image reconstruction and network reliability are considered for this investigation.
Conversely, it is also shown how metaheuristics are very useful for machine learning in determining the optimal neural networks configuration. At this end, new methodologies applied to the multivariate time-series forecasting, and to the inferring gene regulatory networks are presented.
Keywords
Metaheuristics; Machine Learning; Artificial Neural Networks; Optimization Problem; Image Reconstruction; Network reliability Maximization; Hyperparameters Optimization; Multivariate Time-Series Forecasting; Climate Forecasting; Reverse Engineering; Gene Regulatory Network.
Aims and Learning Objectives
The aim of the tutorial is to show how the combination of metaheuristics and machine, i.e., applying one to the other in both ways, can be fruitful in designing fast and reliable algorithms. In light of this, several applications on complex and real-world problems are presented and described.
Target Audience
All conference attendees.
Prerequisite Knowledge of Audience
None.
Detailed Outline
1) introduction to the topic;
2) aims and advantages in using metaheuristics into the machine learning and vice versa;
3) machine learning into GA and IA: image reconstruction, and network reliability as case studies;
4) metaheuristics into machine learning: multivariate time-series forecasting, and gene regulatory networks.
Tutorial on
Structural Bias in Optimization Algorithms
Instructor
|
Anna V. Kononova
Leiden University
Netherlands
|
|
|
|
Brief Bio
Anna V. Kononova is currently an Assistant Professor at the Leiden Institute of Advanced Computer Science (The Netherlands). She received her MSc degree in Applied Mathematics from Yaroslavl State University (Russia) in 2004 and her PhD degree in Computer Science from University of Leeds (UK) in 2010. After 5 years of postdoctoral experience at Technical University Eindhoven (Netherlands) and Heriot-Watt University (Edinburgh, UK), Anna spent 5 years working as an engineer and a mathematician in industry. Her current research interests include analysis of the behaviour of optimisation algorithms and machine learning.
|
Abstract
Abstract
Benchmarking heuristic algorithms is crucial for understanding their performance under different conditions. This tutorial focuses on behaviour benchmarking, specifically addressing the so-called Structural Bias, an inherent bias in iterative heuristic optimizers. By detecting and analyzing Structural Bias, we can enhance algorithm development and identify bias-free conditions. Within this tutorial, we will define Structural Bias, present tools for its detection, discuss computational issues with such detection and visualise examples of structurally biased algorithms. We will showcase the developed toolkits and offer initial insights into mechanisms of the formation of structural bias across various optimization heuristics.
Keywords
optimisation, structural bias, benchmarking, algorithmic behaviour, demo
Aims and Learning Objectives
- convey the importance of benchmarking heuristic algorithms to comprehend their performance across different problem scenarios;
- concentrate on behaviour benchmarking, specifically delving into the concept of Structural Bias (SB) in iterative heuristic optimisers;
- enable participants to detect, analyse, and understand the occurrence and impact of Structural Bias in heuristic optimisation algorithms;
- provide insights into how detecting SB can lead to improved algorithm development and refinement;
- showcase the functionality and usage of developed toolkits as practical tools for detecting and addressing Structural Bias;
- share findings from analysing structural bias across well-known optimisation heuristics, offering insights and patterns.
Target Audience
Researchers who develop heuristic optimisation algorithms or analyse/benchmark such algorithms
Prerequisite Knowledge of Audience
- familiarity with standard heuristic optimisation algorithms and benchmarking practices
- programming examples will be provided in Python
Detailed Outline
Benchmarking heuristic algorithms is vital to understand under which conditions and on what kind of problems certain algorithms perform well. Most benchmarks are performance-based, to test algorithm performance under a wide set of conditions. There are also resource- and behaviour-based benchmarks to test algorithms' resource consumption and behaviour. In this Tutorial, we focus on behaviour benchmarking of algorithms and more specifically we focus on Structural Bias (SB).
SB is a form of bias inherent to the iterative heuristic optimiser in the search space that also affects the performance of the optimisation algorithm. Detecting whether, when and what type of SB occurs in a heuristic optimisation algorithm can provide guidance on what needs to be improved in these algorithms, besides helping to identify conditions under which such bias would not occur.
In the tutorial, we first give the problem definition of detecting and identifying different types of structural bias, including many visual examples. We then introduce different methods for bias detection and demo the BIAS toolkit.
For many well-known and popular optimization heuristics, we have analysed structural bias for different hyper-parameters, we give insights into these results and show best practices to avoid SB in algorithm development.