Zoom link: stanford.zoom.us/meeting/register/

Abstract: Effective decision, exploration, and adaptation often require an agent to know what it knows and, also, what it does not know. This capability relies on the quality of joint predictions of labels assigned to multiple inputs. Conventional neural networks lack this capability and, since most research has focused on marginal predictions, this shortcoming has been largely overlooked. By assessing the quality of joint predictions it is possible to determine whether a neural network effectively distinguishes between epistemic uncertainty (that due to lack of knowledge) and aleatoric uncertainty (that due to chance). We introduce the epistemic neural network (ENN) as a general interface for uncertainty modeling in deep learning. While prior approaches to uncertainty modeling can be viewed as ENNs, the new interface facilitates comparison of joint predictions, and the design of novel architectures and algorithms. In particular, we introduce the epinet: an architecture that can supplement any existing neural network, including pretrained models and be trained with modest incremental computation to represent uncertainty. With an epinet, conventional neural networks outperform very large ensembles, consisting of hundreds or more particles, with orders of magnitude less computation. We demonstrate this efficacy across synthetic data, ImageNet, and sequential decision problems. As part of this effort we open-source experiment code.   

Bio: Ian is a research scientist at DeepMind in Mountain View, working on the design of efficient agents. That means artificial intelligence systems that can learn to perform well in decision problems with a reasonable amount of experience and/or compute. Before DeepMind, Ian completed his PhD at Stanford under the guidance of Benjamin Van Roy, studied Maths at Oxford, and had a brief spell working at JPMorgan in credit derivatives. For more information you can check out his websitehttp://iosband.github.io