No abstract

Figure 1: The proposed method can successfully inpaint large regions and works well with a wide range of images, including those with complex repetitive structures. The method generalizes to high-resolution images, while trained only in low 256 × 256 resolution.

In this work, we investigate Batch Normalization technique and propose its probabilistic interpretation. We propose a probabilistic model and show that Batch Normalization maximazes the lower bound of its marginalized log-likelihood. Then, according to the new probabilistic model, we design an algorithm which acts consistently during train and test. However, inference becomes computationally inefficient. To reduce memory and computational cost, we propose Stochastic Batch Normalization -an efficient approximation of proper inference procedure. This method provides us with a scalable uncertainty estimation technique. We demonstrate the performance of Stochastic Batch Normalization on popular architectures (including deep convolutional architectures: VGG-like and ResNets) for MNIST and CIFAR-10 datasets.

Majority of high-performing off-policy reinforcement learning algorithms use aggregated overestimation bias control techniques.However, most of them rely on a pre-defined bias correction policies that are either not flexible enough or require environment-specific tuning of hyperparameter.In this work, we present a data-driven approach for automatic bias control.We demonstrate its effectiveness on three algorithms: Truncated Quantile Critics, Weighted Delayed DDPG and Maxmin Q-learning. Our approach eliminates the need for an extensive hyperparameter search.We show that it leads to the significant reduction of the actual number of interactions while, in most cases, matching the performance of a resource demanding grid search method.While on average the reduction of the bias improves the performance, elimination of the aggregated bias does not always lead to the best performance. To the best of our knowledge, that is the first case where it is proven on complex environments which highlights the important pitfalls of overestimation control.

No abstract