We cannot guarantee that training datasets are representative of the distribution of inputs that will be encountered during deployment. So we must have confidence that our models do not over-rely on this assumption. To this end, we introduce a new …
Policies trained via Reinforcement Learning (RL) are often needlessly complex, making them more difficult to analyse and interpret. In a run with _n_ time steps, a policy will decide _n_ times on an action to take, even when only a tiny subset of …
Neural networks are vulnerable to adversarially-constructed perturbations of their inputs. Most research so far has considered perturbations of a fixed magnitude under some $l\_p$ norm. Although studying these attacks is valuable, there has been …