Background

The relationship between mind and matter is perhaps the deepest intellectual challenge facing humanity. It links to wide ranges of science, philosophy and engineering, and has been infamously resistant to solution. In recent times, Artificial Intelligence (AI) has become a central theme in addressing this challenge, offering several insights that are both scientifically interesting and practically useful.

Most recently, the field of Deep Neural Networks (DNNs) has brought renewed energy and focus to AI, through a series of remarkable breakthroughs in fields as diverse as speech recognition, board games and self-driving cars. In these and other applications, DNN systems have reached previously unknown levels of accuracy, making human-level performance a distinct possibility and thus suggesting novel insights on the mind-matter problem.

The successes of DNN systems have inspired much research into better algorithms, novel applications and a better understanding of DNNs. The MuST group is involved in all these aspects of DNN research. For example, we are using recurrent DNNs to develop language models for under-resourced languages; these models can be used in tasks such as speech recognition and machine translation. We are also using DNNs in recommendation systems (comparable to those used by Amazon and YouTube to make user-specific suggestions on content and products), and developing acoustic models using these techniques. We balance these applications with theoretical work focussed on understanding and characterising generalisation in the context of deep learning.