We compare the end-to-end training approach to a modular approach in which a system is decomposed into semantically meaningful components. We focus on the sample complexity aspect, in the regime where an extremely high accuracy is necessary, as is the case in autonomous driving applications. We demonstrate cases in which the number of training examples required by the end-to-end approach is exp...

This paper describes a suite of metrics that we use to evaluate our JUPITER system which provides worldwide weather information over the telephone. Since May, 1997, we have made the system available to the general public via a toll-free number and have collected approximately 35,000 utterances to date. These data have proven invaluable for system development and evaluation. Because JUPITER make...

Triangle inequality violations (TIVs) are the effect of packets between two nodes being routed on the longer direct path between them when a shorter detour path through an intermediary is available. TIVs are a natural, widespread and persistent consequence of Internet routing policies. By exposing opportunities to improve the delay between two nodes, TIVs can help myriad applications that seek ...

We have shown, for the first time, that a three component system is capable of aligning gold nanorods (AuNRs) through supramolecular host–guest interactions leading to control over AuNR end-to-end assembly. Viologen end-functionalised AuNRs were prepared that were capable of selectively binding to a cucurbit[8]uril (CB[8]) macrocyclic host molecule. These end-functionalised AuNRs could particip...

In opportunistic networks, end-to-end communication among users does not require a continuous end-to-end path between source and destination. Network protocols are designed to be extremely resilient to events such as long partitions, node disconnections, etc. which are very features of this type of selforganising ad hoc networks. This is achieved by temporarily storing messages at intermediate ...

In this paper we present and discuss our approach to resource management for distributed multimedia applications. We assert that resource management is important to meet the diverse Quality of Service (QoS) needs of multimedia communications in future telecommunications networks, for both the individual communications session and for the network as a whole. Management, in our work, involves mon...

We introduce a novel neural easy-first decoder that learns to solve sequence tagging tasks in a flexible order. In contrast to previous easy-first decoders, our models are end-to-end differentiable. The decoder iteratively updates a “sketch” of the predictions over the sequence. At its core is an attention mechanism that controls which parts of the input are strategically the best to process ne...

Increasing popularity of wireless services has triggered the need for efficient wireless transport mechanisms. TCP, being the reliable transport level protocol widely used in wired network world, was not designed with heterogeneity in mind. The problem with the adaptation of TCP to the evolving wireless settings is because of the assumption that packet loss and unusual delays are mainly caused ...

In wide-area and heterogeneous environments, it is necessary to deploy media gateways at strategic locations within the networks, in order to deliver customized and compose-able multimedia services to individual users. In many cases, media gateways have to be dynamically discovered, because (1) a host may not know a priori all the media gateways it will possibly use, and (2) the change of host ...

We introduce a novel framework for image captioning that can produce natural language explicitly grounded in entities that object detectors find in the image. Our approach reconciles classical slot filling approaches (that are generally better grounded in images) with modern neural captioning approaches (that are generally more natural sounding and accurate). Our approach first generates a sent...