Professor Langford White

I am seeking research consultancy work for up to 10 hours per week in the areas of signal processing, control systems, multi-agent systems, optimisation, game theory. I have extensive experience in working with Defence clients. Please contact me directly for further information.

____________________________________________________________________________________________________________

There are currently two postdoctoral positions (full-time, 2 years potentially renewable) in the area of human-AI interaction funded by the Defence Science and Technology Group. Details are given below. Please contact the technical leads (Prof Ma-Wyatt, A/Prof Nguyen) for additional information.

https://careers.adelaide.edu.au/cw/en/job/505968/postdoctoral-research-fellow-b-school-of-computer-science

https://careers.adelaide.edu.au/cw/en/job/505935/research-fellow-b-with-phd-school-of-psychology

These positions are open to Australian citizens only.

--------------------------------------------------------------------------------------------------------------------------------------------------------------

Professor White's research interests and activities lie in the general areas of statistical signal processing, control, optimisation and multi-agent systems.

Current projects include :

1. Statistical modelling and estimation for hidden reciprocal chains (HRCs)

HRCs are non-causal generalisations of hidden Markov chains and are one-dimensiona Markov random fields, although reciprocal processes are not, in general, Markovian. In this research, we investigate modelling, optimal state estimation, parameter estimation and model classification for HRCs. Applications addressed include "destination-aware" target tracking and the forensic analysis of "pattern-of-life" problems. This work is a collaboration with Dr Francesco Carravetta (IASI-CNR, Italy).

2. Signal subspace methods for sensor array processing of temporally correlated signals

This research is motivated by the question of whether temporal correlation in signals incident of a sensor array can be used to improve source localisation and classification. Advances in computational power mean that previously neglected techniques such as maximum likelihood estimation now become feasible. Recent results concerning the asymptotic consistency of certain estimators (Generalised Statistical Analysis) has also raised the question of whether these approaches can be used in signal subspace based methods such as ESPRIT. This project investigates a range of questions related to these two problems. Collaborators : Prof Peter Sherman (Iowa State Univ., USA), Defence Science and Technology Organisation Australia.

3. (Semi-) Markov Decision Processes, inference and Approximation

This work is primarily concerned with the development of sub-optimal control strategies for very large Markov / Semi-Markov decision processes, where state aggregation or other methods are used to reduce dimensionality. Models considered include Quasi Birth-Death processes and optimally lumped processes. Model estimation from sparse data is a key research challenge.

4. Adaptive games and applications to Communications Systems

In this work, we consider resource allocation in communications networks from a game theoretic point of view. In particular, we examine regret-based reinforcement learning algorithms which, although primarily decentralised, do allow both limited feedback from the resource providers, and the ability for users to form coalitions (i.e. co-operate) to improve performance. These techniques are generic and can be applied in other applications, for example energy markets. Collaboration with Dr Duong Nguyen, the University of Adelaide.

5. Bayesian rationality in satisfaction games

In a satisfaction game, players seek only to attain a satisfactory level of utility rather than competitively maximise their utilities. This research combines the set theoretic specification of satisfaction games and the concept of Bayesian rationality proposed by Robert Aumann (Nobel prize 2005) to develop a new paradigm in game theory. Research is focused on characterisation of equilibria and adaptive algorithms for finding equilibria. Application in resource allocation games are of significance.

Another application domain involves the modelling of the interactions between humans and AI in human-autonomy teaming. In particular, a satisfaction framework is developed using measurable spaces and stopping-time sigma-algebras. Collaboration with Lab_STICC (France) and the School of Psychology, The University of Adelaide.

6. Modelling of asymmetric conflict using Stackelberg games.

Funded by the Defence Science and technology group, this work applies the ideas of Stackelberg games from economics to the problem of modelling asymmetric interactions in coalition games. Such models arise when one set of agents (the leader) has a dominant market power and can for the others to play best-response to its actions. The leader generally improves its utility by announcing its intentions in advance. Model estimation from sparse data is a key problem.

7. Equilibria in search markets

A search market involves agents with different search costs seeking providers for transacting the sale of a good or service. In this research, we address the problem of defining various kinds of equilibrium behaviour and the development of algorithms for finding equilibria. Of particular interest is how sellers can adaptively learn equilibrium strategies through repetition. There is a relationship with Stackelberg games. Collaboration with Prof Roger Waldeck, IMT-Atlantique, France.