This post describes our recent work on Deep Reinforcement Learning (DRL) on the Atari benchmark. DRL appears today as one of the closest paradigm to Artificial General Intelligence and large progress in this field has been enabled by the popular Atari benchmark. However, training and evaluation protocols on Atari vary across papers leading to biased comparisons, difficulty in reproducing results and in estimating the true contributions of the methods. Here we attempt to mitigate this problem with SABER: a Standardized Atari BEnchmark for general Reinforcement learning algorithms. SABER allows us to compare multiple methods under the same conditions against a human baseline and to note that previous claims of superhuman performance on DRL do not hold. Finally, we propose a new state-of-the-art algorithm R-IQN combining Rainbow with Implicit Quantile Networks (IQN). …
This post describes our recent work on unsupervised domain adaptation for semantic segmentation presented at CVPR 2019. ADVENT is a flexible technique for bridging the gap between two different domains through entropy minimization. Our work builds upon a simple observation: models trained only on source domain tend to produce over-confident, i.e., low-entropy, predictions on source-like images and under-confident, i.e., high-entropy, predictions on target-like ones. Consequently by minimizing the entropy on the target domain, we make the feature distributions from the two domains more similar. …
About
