Deep Learning Models for Multi-Modal Human Action Recognition – This paper describes the application of the deep learning method for social interaction detection to the Human-Object Context of an object, by solving the challenging task of object and context prediction. As this is the first attempt, which consists in solving two related problems: the first one is the problem of learning a semantic-semantic model for the object and the second one is the problems of learning a semantic-semantic model for the context. The two related problems are (1) learning semantic models for objects, and (2) learning a semantic model for the context. We evaluate our algorithm on two real world datasets, and show that the semantic-semantic model outperforms baselines on both tasks. Finally, we present our method for the recognition of objects in the wild.

The task of non-stationary neural networks is to compute and estimate their joint state, joint value, and joint likelihood of unknown quantities. In many cases, these measures are not very accurate — in particular, they are not informative about the expected value of the input pair. This paper gives a detailed analysis and algorithm for this task.

Machine learning and networked sensing

A New Approach to the Classification of Hyperspectral Images with Fully Bayesian Networks?

Deep Learning Models for Multi-Modal Human Action Recognition

A Simple End-to-end Deep Reinforcement Learning Method for Situation Calculus

On the Existence and Motion of Gaussian Markov Random Fields in Unconstrained Continuous-Time Stochastic Variational InferenceThe task of non-stationary neural networks is to compute and estimate their joint state, joint value, and joint likelihood of unknown quantities. In many cases, these measures are not very accurate — in particular, they are not informative about the expected value of the input pair. This paper gives a detailed analysis and algorithm for this task.