Sensitivity Analysis of RF+clust for Leave-One-Problem-Out Performance Prediction

2022 IEEE Congress on Evolutionary Computation (CEC), Journal Year: 2023, Volume and Issue: unknown, P. 1 - 8

Published: July 1, 2023

Leave-one-problem-out (LOPO) performance prediction requires machine learning (ML) models to extrapolate algorithms' from a set of training problems previously unseen problem. LOPO is very challenging task even for state-of-the-art approaches. Models that work well in the easier leave-one-instance-out scenario often fail generalize setting. To address problem, recent suggested enriching standard random forest (RF) regression with weighted average on are considered similar test More precisely, this RF+clust approach, weights chosen proportionally distances some feature space. Here work, we extend approach by adjusting distance-based importance features regression. That is, instead considering cosine distance space, consider measure, depending relevance model. Our empirical evaluation modified CEC 2014 benchmark suite confirms its advantages over naive measure. However, also observe room improvement, particular respect more expressive portfolios.

Language: Английский

The Opaque Nature of Intelligence and the Pursuit of Explainable AI

Published: Jan. 1, 2023

In this work We consider and discuss the problems which come with trying to explain human machine intelligence.How explainable artificial intelligence research is being carried out, pitfalls limitations of current approaches bigger question whether we need explanations for trusting inherently complex large intelligent systems, or not.

Language: Английский