This project aims to build novel tools that increase the capacity of art therapists to engage cognitively disabled older people in artistic activities. Engagement with the arts is becoming widely accepted as a method for promoting quality of life in older people. However, many older people have difficulty motivating themselves to engage in a creative activity for a reasonable period of time. These difficulties are compounded when the older adult suffers from a cognitive disability, such as dementia (e.g. Alzheimer's disease).

This project is to design and build real-time systems to assist persons with dementia during handwashing. Assistance is given in the form of verbal and/or visual prompts, or through the enlistment of a human caregiver's help. The system uses only video inputs, and combines a Bayesian sequential estimation framework for tracking hands and towel, with a decision theoretic framework for computing policies of action -- specifically a partially observable Markov decision process (POMDP). A key element of the system is the ability to estimate and adapt to user states, such as awareness, responsiveness and overall dementia level.

This project aims to develop realistic large-scale SSDS with an accurate, extended representation of user goals, and to use new Automatic Belief Compression (ABC) techniques to plan over the large state spaces thus generated. Techniques such as Value-Directed Compression demonstrate that compressible structure can be found automatically in the SSDS domain. Joint work with Oliver Lemon and Paul Crook.

The DyNaMo project investigates probabilistic graphical models (PGMs) and their applications. PGMs allow the representation of probability distributions with many variables in a compact form, also they help to make probabilistic inference (estimate the probability of certain variables given other known) efficiently. PGMs include: Bayesian classifiers, Bayesian networks, Markov random fields, and have many applications in medicine, expert systems, industrial diagnosis, image analysis, robotics and many others.

The Ambient Kitchen is a project at Culture Lab, Newcastle University, run by Patrick Olivier. I collaborate on this project, and am currently looking for a PhD student in machine learning to work with me on it. Email me for details.

This project is developing a real-time system that guides stroke patients during upper extremity rehabilitation. The system automatically modifies exercise parameters to account for the specific needs and abilities of different individuals. The system uses a partially observable Markov decision (POMDP) model of a rehabilitation exercise that can capture this form of customization.

This project uses stereo vision to detect card counters in the game of Blackjack. The system tracks the game as it progresses, monitors the cards played, and tracks the player's betting patterns. The correlation between the player's betting patterns and the game card count is analysed to determine the likelihood that a player is card counting, and the system alerts the Casino staff upon positive identification of a card counter. The Blackjack tracking system also has the ability to detect dealer errors, by monitoring dealers' actions made during the game and ascertaining whether or not the correct action was taken.

SNAP: SyNdetic Assistance Processes for Rapid Development of Technologies for Wellness. This project aims to develop a methodology for rapid prototyping of POMDP models for assistive technology. A POMDP for assistance of persons with dementia has system actions as the various prompts or memory aids the system can use to help a user remember things in the task. For example, shining a light on the kettle may remind them of the kettle's purpose. A key problem is the initial specification of the model for a particular task. We accomplish this by qualitative and quantitative task analysis using Syndetic modeling applied to a small number of example videos. The result is a proces for the initial specification of a full working assistive technology for a particular task. In collaboration with Andrew Monk and Patrick Olivier.

SPUDD stands for Stochastic Planning Using Decision Diagrams. SPUDD implements a value iteration algorithm for MDPs and POMDPs that uses algebraic decision diagrams (ADDs) to represent value functions and policies.