The goal of Bayesian inverse reinforcement learning (IRL) is recovering a posterior distribution over reward functions using a set of demonstrations from an expert optimizing for a reward unknown to the learner. The resulting posterior over rewards can then be used to synthesize an apprentice policy that performs well on the same or a similar task. A key challenge in Bayesian IRL is bridging the computational gap between the hypothesis space of possible rewards and the likelihood, often defined in terms of Q values: vanilla Bayesian IRL needs to solve the costly forward planning problem – going from rewards to the Q values – at every step of the algorithm, which may need to be done thousands of times. We propose to solve this by a simple change: instead of focusing on primarily sampling in the space of rewards, we can focus on primarily working in the space of Q-values, since the computation required to go from Q-values to reward is radically cheaper. Furthermore, this reversion of the computation makes it easy to compute the gradient allowing efficient sampling using Hamiltonian Monte Carlo. We propose ValueWalk – a new Markov chain Monte Carlo method based on this insight – and illustrate its advantages on several tasks.

If autonomous AI systems are to be reliably safe in novel situations, they will need to incorporate general principles guiding them to recognize and avoid harmful behaviours. Such principles may need to be supported by a binding system of regulation, which would need the underlying principles to be widely accepted. They should also be specific enough for technical implementation. Drawing inspiration from law, this article explains how negative human rights could fulfil the role of such principles and serve as a foundation both for an international regulatory system and for building technical safety constraints for future AI systems. This article appears in the AI & Society track.

We point out important problems with the common practice of using the best single model performance for comparing deep learning architectures, and we propose a method that corrects these flaws. Each time a model is trained, one gets a different result due to random factors in the training process, which include random parameter initialization and random data shuffling. Reporting the best single model performance does not appropriately address this stochasticity. We propose a normalized expected best-out-of-$n$ performance ($\text{Boo}_n$) as a way to correct these problems.

Several large cloze-style context-questionanswer datasets have been introduced recently: the CNN and Daily Mail news data and the Children's Book Test.Thanks to the size of these datasets, the associated text comprehension task is well suited for deep-learning techniques that currently seem to outperform all alternative approaches.We present a new, simple model that uses attention to directly pick the answer from the context as opposed to computing the answer using a blended representation of words in the document as is usual in similar models.This makes the model particularly suitable for questionanswering problems where the answer is a single word from the document.Ensemble of our models sets new state of the art on all evaluated datasets.