Robust Semi-Supervised Object Tracking using Conditional Generative Adversarial Networks – Recent studies have shown that the human visual system is well-suited for semantic exploration. We propose a framework for visual exploration based on a new architecture, which can be trained to perform semantic navigation. To our knowledge, this is the first such framework to address the problem of semantic navigation of humans and machines on video. Experimental results validate that deep learning can be used to model human visual exploration on videos of real-world objects in natural settings. This shows that the model can easily scale to thousands of video frames and that the visual exploration ability is at least as rich as deep learning.

We present an algorithm for learning and solving simple logic programs (SMPs) that can be successfully implemented using pure reinforcement learning (RL). This work, called Deep Logic Programming (DLP), is a novel RL technique that aims to harness the state-of-the-art state-of-the-art reinforcement learning methods for reasoning about logic programs. Our approach is based on two simple yet powerful RL tasks: solving the problem of determining the best way to answer a query, and solving the problem of finding a policy based on a random search of a constraint set. We demonstrate that DLP is able to learn to solve complex logic programs using high-dimensional logic programs. We further show that DLP is the best possible option for solving logical programs that do not have any logical properties, and that it is the best available model for reasoning about logic programs that can be learned using purely reinforcement learning methods.

Learning a Hierarchical Bayesian Network Model for Automated Speech Recognition

A Simple Detection Algorithm Based on Bregman’s Spectral Forests

Robust Semi-Supervised Object Tracking using Conditional Generative Adversarial Networks

Augmented Reality at Scale Using Wavelets and Deep Belief Networks

Using Artificial Neurons to Generate Spatial Spaces for Brain-like MachinesWe present an algorithm for learning and solving simple logic programs (SMPs) that can be successfully implemented using pure reinforcement learning (RL). This work, called Deep Logic Programming (DLP), is a novel RL technique that aims to harness the state-of-the-art state-of-the-art reinforcement learning methods for reasoning about logic programs. Our approach is based on two simple yet powerful RL tasks: solving the problem of determining the best way to answer a query, and solving the problem of finding a policy based on a random search of a constraint set. We demonstrate that DLP is able to learn to solve complex logic programs using high-dimensional logic programs. We further show that DLP is the best possible option for solving logical programs that do not have any logical properties, and that it is the best available model for reasoning about logic programs that can be learned using purely reinforcement learning methods.