Convergent Inference Policies for Reinforcement Learning – The goal of this paper is to provide an efficient and robust implementation of a new distributed inference methodology that is able to capture and model the dependencies among agents. We describe the algorithm and the implementation for a new policy architecture, which supports many agents, including many robots. We also discuss the possibility of a future vision for our methodology, which is based on learning to reason.

We propose a new framework for deep learning based feature retrieval from videos, via the use of convolutional neural networks. The purpose is to learn a representation of a video for retrieving important features from a video. In this work, the proposed approach is used on three different datasets, with each dataset being divided into three modules. One module performs features retrieval with the knowledge about the features retrieved from the video. The other module performs feature retrieval with the knowledge about the relevant features retrieved from the video. Experimental results have shown that our approach can generalize to all three modules, and can also lead to accurate retrieval results for both video retrieval and video retrieval of relevant features. The proposed framework is evaluated on three datasets: 1. SVHN dataset, 2. MPII dataset, and 3. Jaccard corpus dataset.

Towards Automated Background Estimation: Recognizing Human Activity in Virtual Artifacts

Deep Reinforcement Learning for Action Recognition

Convergent Inference Policies for Reinforcement Learning

Dynamics from Motion in Images

Image Compression Based on Hopfield Neural NetworkWe propose a new framework for deep learning based feature retrieval from videos, via the use of convolutional neural networks. The purpose is to learn a representation of a video for retrieving important features from a video. In this work, the proposed approach is used on three different datasets, with each dataset being divided into three modules. One module performs features retrieval with the knowledge about the features retrieved from the video. The other module performs feature retrieval with the knowledge about the relevant features retrieved from the video. Experimental results have shown that our approach can generalize to all three modules, and can also lead to accurate retrieval results for both video retrieval and video retrieval of relevant features. The proposed framework is evaluated on three datasets: 1. SVHN dataset, 2. MPII dataset, and 3. Jaccard corpus dataset.