Yoshua Bengio

TL;DR: In this paper, a learned spatial masking mechanism is proposed to filter out irrelevant background signals from attention maps, which results in more accurate localization of discriminatory regions, and the proposed method minimizes mutual information between a masked variational representation and the input while maximizing the information between masked representation and class labels.

TL;DR: It is found that simple architectural choices can outperform existing models in out-of-distribution generalization and show that partitioning relational representations from other information streams may be a simple way to augment existing network architectures’ robustness when performing out- of-dist distribution relational computations.

TL;DR: This work investigates the possibility of empirically establishing performance upper bounds on various visual captioning datasets without extra data labelling effort or human evaluation, and demonstrates the construction of such bounds on MS-COCO, YouTube2Text and LSMDC.

TL;DR: The authors propose to train a backward recurrent network to generate a given sequence in reverse order, and encourage states of the forward model to predict cotemporal states of a backward model.

TL;DR: In this paper, a suite of benchmarking RL environments is designed to evaluate various representation learning algorithms from the literature and find that explicitly incorporating structure and modularity in models can help causal induction in model-based reinforcement learning.