Events defined by the interaction of objects in a scene are often of critical importance; yet important events may have insufficient labeled examples to train a conventional deep model to generalize to future object appearance. Activity recognition models that represent object interactions explicitly have the potential to learn in a more efficient manner than those that represent scenes with global descriptors. We propose a novel inter-object graph representation for activity recognition based on a disentangled graph embedding with direct observation of edge appearance. In contrast to prior efforts, our approach uses explicit appearance for high order relations derived from objectobject interaction, formed over regions that are the union of the spatial extent of the constituent objects. We employ a novel factored embedding of the graph structure, disentangling a representation hierarchy formed over spatial dimensions from that found over temporal variation. We demonstrate the effectiveness of our model on the Charades activity recognition benchmark, as well as a new dataset of driving activities focusing on multi-object interactions with near-collision events. Our model offers significantly improved performance compared to baseline approaches without object-graph representations, or with previous graphbased models.
Abstract-We consider the problem of how to place and efficiently utilize resources in network environments. The setting consists of a regionally organized system which must satisfy regionally varying demands for various resources. The operator aims at placing resources in the regions as to minimize the cost of providing the demands. Examples of systems falling under this paradigm are 1) A peer supported Video on Demand service where the problem is how to place various video movies, and 2) A cloud-based system consisting of regional server-farms, where the problem is where to place various contents or end-user services. The main challenge posed by this paradigm is the need to deal with an arbitrary multi-dimensional (high-dimensionality) stochastic demand. We show that, despite this complexity, one can optimize the system operation while accounting for the full demand distribution. We provide algorithms for conducting this optimization and show that their complexity is pretty small, implying they can handle very large systems. The algorithms can be used for: 1) Exact system optimization, 2) deriving lower bounds for heuristic based analysis, and 3) Sensitivity analysis. The importance of the model is demonstrated by showing that an alternative analysis which is based on the demand means only, may, in certain cases, achieve performance that is drastically worse than the optimal one.
Abstract-TCP has traditionally been considered inappropriate for real-time applications. Nonetheless, popular applications such as Skype use TCP since UDP packets cannot pass through restrictive network address translators (NATs) and firewalls. Motivated by this observation, we study the delay performance of TCP for real-time media flows. We develop an analytical performance model for the delay of TCP. We use extensive experiments to validate the model and to evaluate the impact of various TCP mechanisms on its delay performance. Based on our results, we derive the working region for VoIP and live video streaming applications and provide guidelines for delay-friendly TCP settings. Our research indicates that simple application-level schemes, such as packet splitting and parallel connections, can reduce the delay of real-time TCP flows by as much as 30% and 90%, respectively.
scite is a Brooklyn-based organization that helps researchers better discover and understand research articles through Smart Citations–citations that display the context of the citation and describe whether the article provides supporting or contrasting evidence. scite is used by students and researchers from around the world and is funded in part by the National Science Foundation and the National Institute on Drug Abuse of the National Institutes of Health.