Abstract. While recent supervised dictionary learning methods have attained promising results on the classification tasks, their performance depends on the availability of the large labeled datasets. However, in many real world applications, accessing to sufficient labeled data may be expensive and/or time consuming, but its relatively easy to acquire a large amount of unlabeled data. In this paper, we propose a probabilistic framework for discriminative dictionary learning which uses both the labeled and unlabeled data. Experimental results demonstrate that the performance of the proposed method is significantly better than the state of the art dictionary based classification methods.

A Bayesian framework for 3D human pose estimation from monocular images based on sparse representation (SR) is introduced. Our probabilistic approach aims at simultaneously learning two overcomplete dictionaries (one for the visual input space and the other for the pose space) with a shared sparse representation. Existing SR-based pose estimation approaches only offer a point estimation of the dictionary and the sparse codes. Therefore, they might be unreliable when the number of training examples is small. Our Bayesian framework estimates a posterior distribution for the sparse codes and the dictionaries from labeled training data. Hence, it is robust to overfitting on small-size training data. Experimental results on various human activities show that the proposed method is superior to the stateof-the-art pose estimation algorithms.

No abstract

Distance metric learning is a powerful approach to deal with the clustering problem with side information. For semisupervised clustering, usually a set of pairwise similarity and dissimilarity constraints is provided as supervisory information. Although some of the existing methods can use both equivalence (similarity) and inequivalence (dissimilarity) constraints, they are usually limited to learning a global Mahalanobis metric (i.e., finding a linear transformation). Moreover, they find metrics only according to the data points appearing in constraints, and cannot utilize information of other data points. In this paper, we propose a probabilistic metric learning algorithm which uses information of unconstrained data points (data points which do not appear in neither positive nor negative constraints) along with both positive and negative constraints. We also kernelize our metric learning method based on the kernel trick which provides a non-linear version of the learned metric. Experimental results on synthetic and real-world data sets demonstrate the effectiveness of the proposed metric learning algorithm.

No abstract