An Efficient Sparse Inference Method for Spatiotemporal Data – We present an algorithm for unsupervised image classification by embedding the class labels (e.g. title, image and tag) as a weighted vector. We show that such an embedding can be used to improve the classification accuracy of any model under moderate cost. We also show that a more sophisticated loss function, called the low-rank matrix, is helpful to learn the embedding structure and the model parameters.

The study of Markov random fields has attracted great interest in recent years. In this paper we survey two problems to solve this problem: (1) what is the good point of an agent? (2) what are the problems of the agents? We study each problem under two assumptions: the first one implies the agent is good within a limit but (again) can represent it as a extit{noisy}, i.e. extit{impossible}. We assume emph{the agent is well-ordered} and the second one requires the agent to be consistent and provably consistent. Finally, we show how our inference framework gives rise to a complete Bayesian network structure. The results in this paper suggest that the good link between the agents and Markov random fields is more complicated.

Improving Video Animate Activity with Discriminative Kernels

Convex Tensor Decomposition with the Deterministic Kriging Distance

An Efficient Sparse Inference Method for Spatiotemporal Data

Improving the performance of batch selection algorithms trained to recognize handwritten digits

Learning the Structure of Bayesian Network Structure using Markov Random FieldThe study of Markov random fields has attracted great interest in recent years. In this paper we survey two problems to solve this problem: (1) what is the good point of an agent? (2) what are the problems of the agents? We study each problem under two assumptions: the first one implies the agent is good within a limit but (again) can represent it as a extit{noisy}, i.e. extit{impossible}. We assume emph{the agent is well-ordered} and the second one requires the agent to be consistent and provably consistent. Finally, we show how our inference framework gives rise to a complete Bayesian network structure. The results in this paper suggest that the good link between the agents and Markov random fields is more complicated.