ZhuSuan
In this paper we introduce ZhuSuan, a python probabilistic programming library for Bayesian deep learning, which conjoins the complimentary advantages of Bayesian methods and deep learning. ZhuSuan is built upon Tensorflow. Unlike existing deep learning libraries, which are mainly designed for deterministic neural networks and supervised tasks, ZhuSuan is featured for its deep root into Bayesian inference, thus supporting various kinds of probabilistic models, including both the traditional hierarchical Bayesian models and recent deep generative models. We use running examples to illustrate the probabilistic programming on ZhuSuan, including Bayesian logistic regression, variational auto-encoders, deep sigmoid belief networks and Bayesian recurrent neural networks. …

Trolley Problem
The trolley problem is a thought experiment in ethics. The general form of the problem is this: You see a runaway trolley moving toward five tied-up (or otherwise incapacitated) people lying on the tracks. You are standing next to a lever that controls a switch. If you pull the lever, the trolley will be redirected onto a side track, and the five people on the main track will be saved. However, there is a single person lying on the side track. You have two options:
1. Do nothing and allow the trolley to kill the five people on the main track.
2. Pull the lever, diverting the trolley onto the side track where it will kill one person.
Which is the more ethical option? …

Distance-Based Independence Screening for Canonical Analysis (DISCA)
This paper introduces a new method named Distance-based Independence Screening for Canonical Analysis (DISCA) to reduce dimensions of two random vectors with arbitrary dimensions. The objective of our method is to identify the low dimensional linear projections of two random vectors, such that any dimension reduction based on linear projection with lower dimensions will surely affect some dependent structure — the removed components are not independent. The essence of DISCA is to use the distance correlation to eliminate the ‘redundant’ dimensions until infeasible. Unlike the existing canonical analysis methods, DISCA does not require the dimensions of the reduced subspaces of the two random vectors to be equal, nor does it require certain distributional assumption on the random vectors. We show that under mild conditions, our approach does undercover the lowest possible linear dependency structures between two random vectors, and our conditions are weaker than some sufficient linear subspace-based methods. Numerically, DISCA is to solve a non-convex optimization problem. We formulate it as a difference-of-convex (DC) optimization problem, and then further adopt the alternating direction method of multipliers (ADMM) on the convex step of the DC algorithms to parallelize/accelerate the computation. Some sufficient linear subspace-based methods use potentially numerically-intensive bootstrap method to determine the dimensions of the reduced subspaces in advance; our method avoids this complexity. In simulations, we present cases that DISCA can solve effectively, while other methods cannot. In both the simulation studies and real data cases, when the other state-of-the-art dimension reduction methods are applicable, we observe that DISCA performs either comparably or better than most of them. Codes and an R package can be found in GitHub https://…/DISCA. …

Chargrid
We introduce a novel type of text representation that preserves the 2D layout of a document. This is achieved by encoding each document page as a two-dimensional grid of characters. Based on this representation, we present a generic document understanding pipeline for structured documents. This pipeline makes use of a fully convolutional encoder-decoder network that predicts a segmentation mask and bounding boxes. We demonstrate its capabilities on an information extraction task from invoices and show that it significantly outperforms approaches based on sequential text or document images. …