Department of Computer Science


Artificial Intelligence

profile_image2

Mike Barley (Coordinator)

Adaptive Problem Solving: An automated problem solver can be seen as a combination of algorithms, representations, and heuristics. There are no silver bullets, i.e., no one combination is always better than all others for all problems. My research focusses on developing (1) automated ways to generate new algorithms, new representations, and/or new heuristics; (2) efficient methods to find a good combination of algorithms, representations, and heuristics to solve a given problem; and (3) effective techniques for automatically combining these components into an expert problem solver. The basic idea behind finding a good combination is to use mathematical models of the problem solver that enable it to predict the impact upon its performance of using one combination over another.

Currently, we (Pat Riddle, Santiago Franco, Levi Lellis, Alvaro Torrabla, Carlos Linares Lopez, Daniel Borrajo, Raquel Fuentetaja, Jordan Douglas, and myself) have developed (1) RIDA*, which can automatically generates thousands of heuristics and select a good subset for a given problem when given a representation and algorithm; (2) Baggy, which can automatically generate tens of different representations of a given problem; and (3) MPS, a problem solver, that given a problem, searches the design space over automatically generated representations and heuristics to find a good combination for solving the current problem.  This work has been supported by US AFOSF grants.

As a part of the Adaptive Problem Solving programme, we are exploring heuristic optimal search algorithms.  Our current focus is on bidirectional heuristic search algorithms.  Until recently, it was the case that if strong heuristics were used then A* would be beat all state-of-the-art bidirectional heuristics search algorithms and if the heuristics were poor, then bidirectional blind search would beat them.  We are attempting to create bidirectional heuristic search algorithms that are never worse than either A* or bidirectional blind search regardless of the strength of the heuristic.

Matthew Egbert

How can we create life-like artificial intelligence?

To approach this question, I develop and investigate artificial systems and computational models. Often the systems I study include feedback, where output affects input. So instead of studying how sensor input can be processed to produce desired motor output.

I am more interested in situated, embodied and dynamical approaches where the importance of influence of motors upon sensors is also recognized, and behaviour is seen as the result of the feedback dynamics that operate over the whole sensorimotor loop.

This research follows in the traditions of cybernetics and artificial life, and it contributes to the contemporary enactivist, and dynamical approaches to cognition where the brain, body and world of the agent are all involved in complex feedback relationships and each is playing an essential role in the emergence of intelligent behaviour.

One branch of my research in this area considers how sensorimotor behaviours relate to the origins and early evolution of life. What were the first sensors, motors and behaviours? How did these systems evolve, and how did they influence the evolution of life?

Jiamou Liu

A multi-agent system (MAS) consists of multiple interacting agents who function autonomously within an environment. My research focuses on MAS that involves agents who form a social structure, and the analysis of this social structure through social network analysis and game theory. Topics of interests include interpersonal ties, information diffusion, communities structures, link formation, influence maximisation, and social modelling and simulation.

Pat Riddle

I have a broad area of research interest.  I have worked mainly in Machine Learning over the years.  But currently I am also working in automated problem solving and planning.  Specifically I work on domain modeling and problem reformulation to automatically derive new representations which are easier for the problem solver to use. I also am working in a new area for me: on heuristic optimal search.

In machine learning I have worked with many different algorithms.  In all the areas, probably one of two main issues occur: either we are trying to understand why the machine learning algorithm behaves the way it does OR we are trying to use the algorithm in a novel way to solve a specific problem.

Ian Watson

My principle areas of research are, case-based reasoning (CBR), applied artificial intelligence and knowledge management. I am one of the most prominent researchers in the CBR community. Until recently I was on the Technical Advisory Board of Inference Corporation, a US company that pioneered the commercialisation of CBR (recently taken over by eGain www.egain.com) and have been a consultant to Orion (NZ), AMEC Engineering Ltd., Unisys, Legal & General, the Royal Automobile Club, and the UK Government’s Cabinet Office, winning the UKGovernment Computing Award for Innovation.

In recent years I have applied CBR and machine learning to computer games (strategy games like Civilization, StarCraft and Texas Hold’em Poker). Our poker playing agent competes in the annual Computer Poker Competition placing well. I have also established a Game AI research group (www.cs.auckland.ac.uk/reasearch/gameai/) and have established close links with the NZ game development community including consulting for SmallWorlds(www.smallworlds.com).

Members (academic staff or PhD students):

Academic staff:

Mike Barley, Matthew Egbert, Jiamou LiuPat Riddle, Ian Watson

Phd students:

Rehab Alkhater, Bella Anderson, Diana Benavides PradoFeiyu (Albert) HuSaima Tariq KhanAlex PengShen SongRalph Versteegan, Terry Marange, Devipriyaa Nagappan