School of
Information Technology and Electrical Engineering

Speaker: Joshua Arnold
Seminar Date: Mon, 09/10/2017 - 14:00
Venue: 47A-249 (Sir James Foots)
Host: Prof Janet Wiles

Seminar Type:  PhD Confirmation Seminar


An artificial agent acting or sensing in the real world must have a system capable of encoding time and its actions together either implicitly or explicitly. Often, time is implicit within a model and overlooked for the complications it adds to a task. This is the case in many neural models which discretise time into uniform steps at which to sample the world. Recently, there has been a surge of interest in neuromorphic sensors which asynchronously sample the world like the biological sensors of mammals. This has facilitated a paradigm shift where time directly influences computation. This work will analyse the prediction performance of neural networks with different data representations that make temporal information explicit. The representations and models will then be validated in a real world task of a ‘curious’ robot that actively explores phenomena that it can only partially predict in an attempt to improve its model. Results will demonstrate the performance possible with explicit temporal representations and show they can be functional in real world tasks.


Joshua Arnold received his Bachelors of Engineering (Software) from The University of Queensland in 2016. He is currently a PhD student under the supervision of Prof. Janet Wiles.