Probabilistic programs in high-dimensional domains – We present a principled framework for learning a probabilistic programming language by probabilistic programming. This framework takes as input a probabilistic programming grammar and learns a parser from this grammar. These grammar parsers are known to be good at probabilistic programming, and we propose a language for learning parsers that can learn parsers from probabilistic programs. We show that this parsers can be trained efficiently on synthetic and real data, and the framework is robust to the constraints imposed by the real data. We also report our analysis of the learning task of the parser for probabilistic programs.

The recently proposed deep network algorithms have shown remarkable ability to achieve state of the art performance on the task of video object classification. In addition to providing state of the art performance, these algorithms also offer a novel and yet challenging task for human participants. Despite our best efforts, learning of the deep network is still a very challenging task for human users. Despite the fact that deep architectures have been widely used for this task, the performance of convolutional neural networks (CNNs) is still very much dominated by network-based tasks. In this work, we aim to establish a new benchmark for CNN learning.

Auxiliary Reasoning (OBWK)

A Novel Architecture for Building Datasets of Constraint Solvers

Probabilistic programs in high-dimensional domains

Fast Non-convex Optimization with Strong Convergence Guarantees

Efficient Construction of Deep Neural Networks Using Conditional Gradient and SparsityThe recently proposed deep network algorithms have shown remarkable ability to achieve state of the art performance on the task of video object classification. In addition to providing state of the art performance, these algorithms also offer a novel and yet challenging task for human participants. Despite our best efforts, learning of the deep network is still a very challenging task for human users. Despite the fact that deep architectures have been widely used for this task, the performance of convolutional neural networks (CNNs) is still very much dominated by network-based tasks. In this work, we aim to establish a new benchmark for CNN learning.