- 156 views
- 228 downloads
Detecting Anomalous Collective Behaviours in Simulated Multi-Agent Environments
-
- Author / Creator
- Schumacker Soares, Everton
-
Several Artificial Intelligence (AI) techniques such as machine learning, evolutionary computing, and Artificial Life (A-life) have been increasingly used to generate emergence of novel behaviours in multi-agent simulations (e.g., commercial games). However, automatically detecting emergent behaviours and recognizing which ones are indeed novel/anomalous still poses a challenging problem. The solution for such a problem may potentially help lead to improvements in quality assurance control (e.g., detection of bugs) and security (e.g., detection of suspicious behaviour in surveillance footage). Some of previously published attempts to detect anomalous behaviour in simulations relied on machine learning techniques. For example, the use of supervised learning models to detect anomalous behaviours requires labelled data not always available at the training time due to the rarity of anomalous behaviours. On another hand, the use of unsupervised learning models usually relies on specific types of behaviour patterns or on having access to internal simulation states. Our approach presented in this thesis uses only unlabelled sequences of readily available visualization frames from simulated multi-agent environments to train a deep variational autoencoder (i.e., a deep artificial neural network). After being trained, the autoencoder can detect anomalous behaviours by comparing its reconstruction error against a threshold. We tested our approach in a predator/prey A-life environment where it proved viable, being even robust to a certain amount of pollution (i.e., the inclusion of anomalous data) in the training data. As a case study, we also applied our approach to a video game where the results are yet inconclusive.
-
- Graduation date
- Spring 2020
-
- Type of Item
- Thesis
-
- Degree
- Master of Science
-
- License
- Permission is hereby granted to the University of Alberta Libraries to reproduce single copies of this thesis and to lend or sell such copies for private, scholarly or scientific research purposes only. Where the thesis is converted to, or otherwise made available in digital form, the University of Alberta will advise potential users of the thesis of these terms. The author reserves all other publication and other rights in association with the copyright in the thesis and, except as herein before provided, neither the thesis nor any substantial portion thereof may be printed or otherwise reproduced in any material form whatsoever without the author's prior written permission.