CDT Pizza 30/11

Talk #1

Speaker

Andrew Brock

Title

BigGAN: Large Scale GAN Training for High Fidelity Natural Image Synthesis

Abstract

Despite recent progress in generative image modeling, successfully generating high-resolution, diverse samples from complex datasets such as ImageNet remains an elusive goal. In this talk, I'll give an overview of my internship work at DeepMind, training Generative Adversarial Networks at the largest scale yet attempted, and studying the instabilities specific to such scale. The modifications we propose lead to a 200-300% improvement in the state of the art in conditional image synthesis (as measured by Inception Score and FID), and allow us to finely control the tradeoff between sample quality and diversity.

Talk #2

Speaker

Alan Bundy

Title

Darpa's Third Wave of AI

Abstract

The US funding agency DARPA is investing $2Bn in a research programme to combine the best of symbolic and sub-symbolic AI. It's slogan is that "Systems [will] construct contextual explanatory models for classes of real world phenomena". It divides the history of AI into 3 waves: the first wave was symbolic AI, e.g., expert systems; the 2nd wave is statistical machine learning; and the 3rd wave will combine these. Edinburgh has strengths in the first two waves. We should examine whether we can profitably join forces.