Unsupervised Video Summarization via Deep Learning – Video synthesis has been proposed as a technique to improve the performance of a video synthesis task. In this paper, we investigate the effect of several recent video synthesis methods on video synthesis tasks. We study two different video synthesis methods using an adversarial framework to generate video frames with different levels of classification. First, we propose an unsupervised classifier called VideoNet-AUC to generate low-level classification frames. In addition, we propose a method to predict visual attributes such as color, texture, and size. We demonstrate the effectiveness of the proposed method on three publicly available datasets and compare the results. The proposed method compared favorably with the unsupervised methods on multiple video synthesis tasks.

We present a framework for solving a generalised non-convex, non-linear optimization problem where the objectives are to efficiently recover a solution to a constraint, and the solutions are generated by an approximate search algorithm. The algorithms we describe are generalised to the standard PC solvers and provide a generalisation of these algorithms to the non-convex case. We provide an algorithm description for the standard PC solver, which is based on a non-convex optimization problem and a constraint solver, namely the Non-Zero Satisfiability Problem (NSSP). Based on the proposed algorithm, we illustrate how it can be used on general convex optimization problems with an objective function that is guaranteed to be linear in the solution dimensions. Our main result is that the algorithm has a reasonable guarantee of solving any constraint whose objective function is a non-convex. We also illustrate how to use any constraint solver to compute the solution to a non-convex optimization problem with a constraint objective function.

Makeshift Dictionary Learning on Discrete-valued Texture Pairings

Linear Sparse Coding via the Thresholding Transform

Unsupervised Video Summarization via Deep Learning

Bayesian Networks in Naturalistic Reasoning

Lifted Dynamical Stochastic Complexity (LSPC): A Measure of Difference, Stability, and SimilarityWe present a framework for solving a generalised non-convex, non-linear optimization problem where the objectives are to efficiently recover a solution to a constraint, and the solutions are generated by an approximate search algorithm. The algorithms we describe are generalised to the standard PC solvers and provide a generalisation of these algorithms to the non-convex case. We provide an algorithm description for the standard PC solver, which is based on a non-convex optimization problem and a constraint solver, namely the Non-Zero Satisfiability Problem (NSSP). Based on the proposed algorithm, we illustrate how it can be used on general convex optimization problems with an objective function that is guaranteed to be linear in the solution dimensions. Our main result is that the algorithm has a reasonable guarantee of solving any constraint whose objective function is a non-convex. We also illustrate how to use any constraint solver to compute the solution to a non-convex optimization problem with a constraint objective function.