Damian Borowiec

Damien Borowiec’s thesis was awarded 30/06/2023. ‘Analysing and Reducing Costs of Deep Learning Compiler Auto-tuning’ examined the AI technique Deep Learning (DL) and how to understand its social and environmental costs. DL is significantly impacting many aspects of society, being used by industry, in medicine, and increasingly in consumer products. But DL systems are increasingly complex, and their computational loads can consume increasingly large amounts of energy. In his thesis, Borowiec reports on how he devised and investigated two systems for the better prediction of these costs, showing how computational value – a material value – can be placed alongside a social value, namely the benefits a DL application offers a user. Three major publications derived, two on the applications he built, and a third on the general question of sustainable computing. Damien says of his research: ‘My experience with MSF research expanded my perspective on the broader societal and environmental implications of computing and AI in particular. With this background, I have felt more able to engage with real world problems such as those considered by my first full time employer, a company that seeks to predict and calculate completion times and costs of large-scale construction projects. Looking ahead, I aim to lead research into innovative AI technologies that address the environmental consequences of AI usage.’