Faculty/School

Faculty of Engineering

School of Electrical Engineering and Robotics

Topic status

We're looking for students to study this topic.

Research centre

Centre for Robotics

Supervisors

Professor Will Browne
Position
Professor and Chair in Manufacturing Robotics
Division / Faculty
Faculty of Engineering

Overview

AI that is pre-programmed is limited on its tasks and human bias. Learning systems offer richer behaviour, where collaborators and I have created the following that need developing (projects bulleted)

A symbolic system that works in Boolean and Integer domains without noise

  • Need to translate this to real-valued, noisy domains. Code exists, but the data sets do not and integration is still uncertain. Most intriguing is how future use can influence current learning, i.e. "will this be useful in the future?"

A lateralized system exists that enables an input to be considered at the local level and the holistic level simultaneously.

  • How lateralization functions in continual learning is an open question, e.g. can we learn the concept of a leg, wheel, etc., so that we can readily identify and differentiate forms of locomotion?

A compaction algorithm exists that can remove redundant and irrelevant learnt knowledge, where it is unknown how this works for continual learning. What happens with almost duplicate knowledge from different sources, how to forget (without catastrophic forgetting), how to learn economically and so forth.

  • How to learn in a distributed/federated manner continuously is an open question, but recent advances in networked learning are promising.

Research engagement

Literature review to up skill on the latest aspects of machine learning.

Methods for continual learning within artificial intelligence systems

Research activities

Computer programming of symbolic and/or connectionist artificial intelligence. Potential languages include Python, and ideally Julia.
Experimental testing of ideas on artificial datasets, such as abstraction reasoning corpus, with the expansion to robotic systems possible.

Outcomes

Computer program with improved levels of artificial intelligence when addressing complex problems.

Skills and experience

Strong interest and background in computer programming, including artificial intelligence. Programming in Julia is an advantage.

Start date

1 November, 2024

End date

14 January, 2025

Location

S-Block Gardens Point, liekly S11

Additional information

An existing doctorate student, Cameron, will be able to provide advice on the abstraction reasoning corpus and methods used to address this problem.

Keywords

Contact

Will Browne, will.browne@qut.edu.au