Highlighting spatiotemporal patterns in time series with CNNs – We present the first deep CNN, which incorporates multiple layers of CNNs into a single layer per network. Through multiple layers, we utilize multilayers to learn the structure of the data structure, and use the structure of multilayers as a pre-processing step to refine the CNN. Experiments on datasets of 50,000 users show the superiority of the proposed model, which is much faster than traditional CNN approaches by orders of magnitude.
We present a simple but powerful feature descriptor for the feature extraction of images in an unsupervised setting. We first show how to make use of the descriptor to extract important information about a subject, e.g. whether it are a bird or a dog. We then propose a method to retrieve the information from images by performing a pre-defined sequence of feature extraction steps. The proposed descriptor is capable of retrieving information about the object in the images, by using a different type of filter. We present experiments on the KITTI dataset, a set of 15 annotated images from around the world, highlighting how the descriptor could help in the extraction of information from images.
This paper proposes a novel method based on the use of probabilistic inference and supervised learning for learning a Bayesian network from a Bayesian network. Given parameters and a conditional model, the goal is to find a posterior distribution that is of interest in the learning process. In particular, it is required that the posterior can be found in a structured environment. As in the Bayesian model, the posterior is constructed from the set of constraints that are relevant to the learner’s expected utility function for the learner, and the knowledge that the learner may have for the learner by using a prior.
A deep learning algorithm for removing extraneous features in still images
An Analysis of the Determinantal and Predictive Lasso
Highlighting spatiotemporal patterns in time series with CNNs
On the convergence of the gradient of the Hessian
Learning Hierarchical Networks through Regularized Finite-Time UpdatesThis paper proposes a novel method based on the use of probabilistic inference and supervised learning for learning a Bayesian network from a Bayesian network. Given parameters and a conditional model, the goal is to find a posterior distribution that is of interest in the learning process. In particular, it is required that the posterior can be found in a structured environment. As in the Bayesian model, the posterior is constructed from the set of constraints that are relevant to the learner’s expected utility function for the learner, and the knowledge that the learner may have for the learner by using a prior.