| |||||||
Technical Keynote Lecture
Prof. Marc Cavazza (University of Greenwich)
The Unreasonable Effectiveness of Non-admissible SearchAbstract The concept of admissibility in heuristic search is one of the best known in the entire field of Artificial Intelligence. Although formal studies of heuristics have established quite early that a controlled departure from admissibility made it possible to trade optimality for performance, the past ten years have seen a growing interest in non-admissible heuristics for various AI applications, ranging from multi-agent search to heuristic search planning. Beyond the performance implications of non-admissible heuristics, one lesser-known aspect consists in the representational ability of non-admissible search methods, which can support the incorporation of application knowledge to guide the search process. This talk will introduce recent examples in the use of non-admissible heuristic search in real-world or complex problems remote from traditional benchmarks. These will include the incorporation of application data, such as users’ preferences, in route planning tasks, as well as recent work in Multi-Agent Path Finding to develop traffic management systems for Unmanned Aerial Vehicles. Finally, we will discuss the use of non-admissible search in Human Augmentation or hybrid AI systems as a mechanism allowing experiments controlling heuristic search from humans’ motivational dispositions, via Brain-Computer Interfaces.
Marc Cavazza is Professor and Head of the School of Computing and Mathematical Sciences at the University of Greenwich. He started his career in AI and Medicine and has later conducted research in intelligent user interfaces and intelligent media, primarily applying planning techniques to narrative generation. His work on narrative technology has been funded by several EU grants, including one Network of Excellence and two FET projects. He has published at multiple AI venues such as AAMAS, IJCAI, AIME, ECAI and ICAPS, and received the best demonstration award at AAMAS 2010 and ICAPS 2013.
| |||||||
