IPAB Workshop - 30/01/2020

Title: Deep Learning: Better, Faster, Stronger

Abstract:

Deep Neural Networks are arguably near the “peak of inflated expectations” in the technology hype cycle. Success in applications from vision to speech and game playing, have led to big commercial investments. However issues including their vast data and energy requirements and vulnerability to domain-shift, out-of-bounds generalisation, and adversarial attack makes their place in our future unclear. The “trough of disillusionment” looms. 

In this talk I will provide an overview of my group’s research in this area -- aiming to make DNNs real-world ready in terms of data efficiency, robustness and generalisation through building in prior knowledge, improving learning algorithms with meta-learning, and neuro-symbolic learning and reasoning.