Research Project

Machine Learning and Intelligent Agents Logo and Link

Machine Learning and
Intelligent Agents

Interpretable Models for Intelligent Data Analysis & System Identification


Dr. John Gan

Project Description:

With the ever-increasing quantity of electronic data available from complex industrial and commercial processes and systems whose first-principle models are unknown, it is highly desirable to construct data-driven models that have the capability to interpret data and to predict the future behaviours of these processes and systems. One challenging problem occurs in this area when the input space dimension of the system to be modelled is high. In this situation, it is extremely difficult to model the complex system accurately using a single global model, and it is also very challenging to use local modelling approaches, such as advanced fuzzy modelling techniques, due to the curse of dimensionality problem. Another interesting and challenging problem with local modelling is how to achieve not only a good global approximation but also close approximations of local linearisations of the nonlinear system by local models, as this requirement will enhance the interpretability and applicability of the local models. We are investigating criteria for model interpretability, and developing new local modelling techniques and the relevant theory for generating interpretable models from data.



  Tuesday, 26 October 2004