Talks and presentations

AI for Expensive Optimization Problems in Industry

June 05, 2023

poster, CAI 2023, Santa Clara, California, United States

The optimization of real-world engineering problems can be a challenging task, due to the limited understanding of problem characteristics and the high computational cost of objectives and constraints. This study proposes an AI-assisted optimization pipeline that addresses these challenges by using proxy functions in order to select and optimize an optimization algorithm and its hyper-parameters. It thereby significantly accelerates the optimization process on the real (expensive) problem. To obtain such proxy functions Exploratory Landscape Analysis (ELA) features are used to characterize the problem’s landscape. The ELA features are then used to identify an artificial function that replicates the original problem’s properties.

Deep Learning, A broad introduction

July 30, 2019

Invited talk, Ecole Summer school, Leiden, The Netherlands

Invited talk for the Summer school for Early Stage Researchers. This summer school takes place within the EU project Experience-based Computation: Learning to Optimise (ECOLE), which investigates novel synergies between nature-inspired optimisation and machine learning to address key challenges faced by the European industry.

Datamining & Apps

September 23, 2016

Invited guest lecture, Data Science course, Liacs, Leiden, The Netherlands

Invited lecture at the Data Science course, LIACS, Leiden University

Optimally Weighted Cluster Kriging

October 22, 2015

talk, IDA 2015, Saint-Etienne, France

In business and academia we are continuously trying to model and analyze complex processes in order to gain insight and optimize. One of the most popular modeling algorithms is Kriging, or Gaussian Processes. A major bottleneck with Kriging is the amount of processing time of at least O(n3) and memory required O(n2) when applying this algorithm on medium to big data sets. With big data sets, that are more and more available these days, Kriging is not computationally feasible. As a solution to this problem we introduce a hybrid approach in which a number of Kriging models built on disjoint subsets of the data are properly weighted for the predictions. The proposed model is both in processing time and memory much more efficient than standard Global Kriging and performs equally well in terms of accuracy. The proposed algorithm is better scalable, and well suited for parallelization.