Abstract Large Language Models (LLMs) are often aligned using contrastive alignment objectives and preference pair datasets. The interaction between model, paired data, and objective makes alignment a complicated procedure, sometimes producing subpar results. We study this and find that (i) preference data gives a better learning signal when the underlying responses are contrastive, and (ii) alignment objectives lead to better performance when they specify more control over the model during training. Based on these insights, we introduce Contrastive Learning from AI Revisions (CLAIR), a data-creation method which leads to more contrastive preference pairs, and Anchored Preference Optimization (APO), a controllable and more stable alignment objective. We align Llama-3-8B-Instruct using various comparable datasets and alignment objectives and measure MixEval-Hard scores, which correlate highly with human judgments. The CLAIR preferences lead to the strongest performance out of all datasets, and APO consistently outperforms less controllable objectives. Our best model, trained on 32K CLAIR preferences with APO, improves Llama-3-8B-Instruct by 7.65%, closing the gap with GPT4-turbo by 45%. Our code and datasets are available.

Abstract In the development of a task-oriented dialogue system, defining the dialogue structure is a time-consuming task. Hence, several works have looked into automatically inferring it from data, e.g., actual conversations between a customer and a support agent. To recover such dialogue structure, recent methods based on discrete variational models learn to jointly encode and cluster utterances in dialogue states, but (i) represent utterances by only considering preceding dialogue context, and (ii) are slow to train since they are optimized with a compute-expensive decoding objective. We revisit and improve upon an existing efficient pipeline approach, commonly adopted as a baseline, that first encodes utterances and then clusters them with k-means to induce the dialogue structure. However, the existing approach represents utterances as bag-of-words or skip-thought vectors, which have been shown to perform poorly in semantic similarity tasks, and without considering dialogue context. We therefore first investigate the use of more powerful transformer-based encoders for encoding utterances. Next, we propose ellodar, a method for learning representations that capture both preceding and subsequent dialogue context, inspired by word-to-vec training strategies. ellodar is efficient since representations are learned directly in the encoding space by finetuning just a single linear layer on top of a frozen sentence encoder with a vector-to-vector regression training objective. Extensive experiments on representative datasets for dialogue structure induction (SimDial, Schema Guided Dialogues, DSTC2, and CamRest676) demonstrate that in terms of effectiveness to induce the correct dialogue structure, (i) clustering utterances represented by transformed-based encoders improves recent joint models by 13%–32% on standard cluster metrics, and (ii) clustering ellodar’s representations yields additional improvements ranging from +20% to +26%, with speedups of $$\times $$ × $$\textbf{10}$$ 10 – $$\textbf{10}^{\textbf{4}}$$ 10 4 compared to the recent joint models.

In energy-intensive industrial systems, an evaporative cooling process may introduce operational flexibility. Such flexibility refers to a system’s ability to deviate from its scheduled energy consumption. Identifying the flexibility, and therefore, designing control that ensures efficient and reliable operation presents a great challenge due to the inherently complex dynamics of industrial systems. Recently, machine learning (ML) models have attracted attention for identifying flexibility, due to their ability to model complex nonlinear behavior. This article presents ML-based methods that integrate system dynamics into the ML models (e.g., neural networks) for better adherence to physical constraints. We define and evaluate physics-informed long-short term memory networks (PhyLSTMs) and physics-informed neural networks (PhyNN) for the identification of flexibility in the evaporative cooling process. These physics-informed networks approximate the time-dependent relationship between control input and system response while enforcing the dynamics of the process in the neural network architecture. Our proposed PhyLSTM provides less than 2% system response estimation error, converges in less than half iterations compared to a baseline NN, and accurately estimates the defined flexibility metrics. We include a detailed analysis of the impact of training data size on the performance and optimization of our proposed models.

For text classification tasks, finetuned language models perform remarkably well.Yet, they tend to rely on spurious patterns in training data, thus limiting their performance on outof-distribution (OOD) test data.Among recent models aiming to avoid this spurious pattern problem, adding extra counterfactual samples to the training data has proven to be very effective.Yet, counterfactual data generation is costly since it relies on human annotation.Thus, we propose a novel solution that only requires annotation of a small fraction (e.g., 1%) of the original training data, and uses automatic generation of extra counterfactuals in an encoding vector space.We demonstrate the effectiveness of our approach in sentiment classification, using IMDb data for training and other sets for OOD tests (i.e., Amazon, SemEval and Yelp).We achieve noticeable accuracy improvements by adding only 1% manual counterfactuals: +3% compared to adding +100% in-distribution training samples, +1.3% compared to alternate counterfactual approaches.

Soft actuators offer a safe, adaptable approach to tasks like gentle grasping and dexterous manipulation. Creating accurate models to control such systems however is challenging due to the complex physics of deformable materials. Accurate Finite Element Method (FEM) models incur prohibitive computational complexity for closed-loop use. Using a differentiable simulator is an attractive alternative, but their applicability to soft actuators and deformable materials remains under-explored. This paper presents a framework that combines the advantages of both. We learn a differentiable model consisting of a material properties neural network and an analytical dynamics model of the remainder of the manipulation task. This physics-informed model is trained using data generated from FEM, and can be used for closed-loop control and inference. We evaluate our framework on a dielectric elastomer actuator (DEA) coin-pulling task. We simulate the task of using DEA to pull a coin along a surface with frictional contact, using FEM, and evaluate the physics-informed model for simulation, control, and inference. Our model attains ≤ 5% simulation error compared to FEM, and we use it as the basis for an MPC controller that requires fewer iterations to converge than model-free actor-critic, PD, and heuristic policies.

In our continuously evolving world, entities change over time and new, previously non-existing or unknown, entities appear. We study how this evolutionary scenario impacts the performance on a well established entity linking (EL) task. For that study, we introduce TempEL, an entity linking dataset that consists of time-stratified English Wikipedia snapshots from 2013 to 2022, from which we collect both anchor mentions of entities, and these target entities’ descriptions. By capturing such temporal aspects, our newly introduced TempEL resource contrasts with currently existing entity linking datasets, which are composed of fixed mentions linked to a single static version of a target Knowledge Base (e.g., Wikipedia 2010 for CoNLL-AIDA). Indeed, for each of our collected temporal snapshots, TempEL contains links to entities that are continual, i.e., occur in all of the years, as well as completely new entities that appear for the first time at some point. Thus, we enable to quantify the performance of current state-of-the-art EL models for: (i) entities that are subject to changes over time in their Knowledge Base descriptions as well as their mentions’ contexts, and (ii) newly created entities that were previously non-existing (e.g., at the time the EL model was trained). Our experimental results show that in terms of temporal performance degradation, (i) continual entities suffer a decrease of up to 3.1% EL accuracy, while (ii) for new entities this accuracy drop is up to 17.9%. This highlights the challenge of the introduced TempEL dataset and opens new research prospects in the area of time-evolving entity disambiguation.

Powerful sentence encoders trained for multiple languages are on the rise.These systems are capable of embedding a wide range of linguistic properties into vector representations.While explicit probing tasks can be used to verify the presence of specific linguistic properties, it is unclear whether the vector representations can be manipulated to indirectly steer such properties.For efficient learning, we investigate the use of a geometric mapping in embedding space to transform linguistic properties, without any tuning of the pre-trained sentence encoder or decoder.We validate our approach on three linguistic properties using a pre-trained multilingual autoencoder and analyze the results in both monolingual and crosslingual settings.

Amir Hadifar, Sofie Labat, Veronique Hoste, Chris Develder, Thomas Demeester. Proceedings of the 2021 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies. 2021.

Inducing sparseness while training neural networks has been shown to yield models with a lower memory footprint but similar effectiveness to dense models.However, sparseness is typically induced starting from a dense model, and thus this advantage does not hold during training.We propose techniques to enforce sparseness upfront in recurrent sequence models for NLP applications, to also benefit training.First, in language modeling, we show how to increase hidden state sizes in recurrent layers without increasing the number of parameters, leading to more expressive models.Second, for sequence labeling, we show that word embeddings with predefined sparseness lead to similar performance as dense embeddings, at a fraction of the number of trainable parameters.

Adversarial training (AT) is a regularization method that can be used to improve the robustness of neural network methods by adding small perturbations in the training data.We show how to use AT for the tasks of entity recognition and relation extraction.In particular, we demonstrate that applying AT to a general purpose baseline model for jointly extracting entities and relations, allows improving the state-of-the-art effectiveness on several datasets in different contexts (i.e., news, biomedical, and real estate data) and for different languages (English and Dutch).

Non-intrusive load monitoring detects active appliances in a household (and their power consumption) from measuring the aggregated power at just one point in that household. Our previous works focused on classifying a single appliance, assuming that the voltage and current trace could be isolated from an aggregated signal by considering the difference in current before and after the event. In this paper, we show that this assumption holds and that it is a viable approach in practice. We experimentally validate this for two classification methods we proposed earlier: (1) random forests using elliptical Fourier descriptors of the appliances' VI trajectories and (2) convolutional neural networks using the appliances' VI images. We benchmark these approaches on the aggregated data from the 2018 version of PLAID. We obtain, respectively for each of these classifiers, a maximal Fmacro-measure of 85.31% and 87.95%. We also show that using submetered data for training does not improve the performance.

Comprehending lyrics, as found in songs and poems, can pose a challenge to human and machine readers alike. This motivates the need for systems that can understand the ambiguity and jargon found in such creative texts, and provide commentary to aid readers in reaching the correct interpretation. We introduce the task of automated lyric annotation (ALA). Like text simplification, a goal of ALA is to rephrase the original text in a more easily understandable manner. However, in ALA the system must often include additional information to clarify niche terminology and abstract concepts. To stimulate research on this task, we release a large collection of crowdsourced annotations for song lyrics. We analyze the performance of translation and retrieval models on this task, measuring performance with both automated and human evaluation. We find that each model captures a unique type of information important to the task.

The problem of noisy and unbalanced training data for supervised keyphrase extraction results from the subjectivity of keyphrase assignment, which we quantify by crowdsourcing keyphrases for news and fashion magazine articles with many annotators per document.We show that annotators exhibit substantial disagreement, meaning that single annotator data could lead to very different training sets for supervised keyphrase extractors.Thus, annotations from single authors or readers lead to noisy training data and poor extraction performance of the resulting supervised extractor.We provide a simple but effective solution to still work with such data by reweighting the importance of unlabeled candidate phrases in a two stage Positive Unlabeled Learning setting.We show that performance of trained keyphrase extractors approximates a classifier trained on articles labeled by multiple annotators, leading to higher average F 1 scores and better rankings of keyphrases.We apply this strategy to a variety of test collections from different backgrounds and show improvements over strong baseline models.

Information extraction (IE) systems discover structured information from natural language text, to enable much richer querying and data mining than possible directly over the unstructured text. Unfortunately, IE is generally a computationally expensive process, and hence improving its efficiency, so that it scales over large volumes of text, is of critical importance. State-of-the-art approaches for scaling the IE process focus on one text collection at a time. These approaches prioritize the extraction effort by learning keyword queries to identify the "useful" documents for the IE task at hand, namely, those that lead to the extraction of structured "tuples." These approaches, however, do not attempt to predict which text collections are useful for the IE task---and hence merit further processing---and which ones will not contribute any useful output---and hence should be ignored altogether, for efficiency. In this paper, we focus on an especially valuable family of text sources, the so-called deep web collections, whose (remote) contents are only accessible via querying. Specifically, we introduce and study techniques for ranking deep web collections for an IE task, to prioritize the extraction effort by focusing on collections with substantial numbers of useful documents for the task. We study both (adaptations of) state-of-the-art resource selection strategies for distributed information retrieval, and IE-specific approaches. Our extensive experimental evaluation over realistic deep web collections, and for several different IE tasks, shows the merits and limitations of the alternative families of approaches, and provides a roadmap for addressing this critically important building block for efficient, scalable information extraction.

To realize the Internet of Things (IoT) vision, tools are needed to ease the development and deployment of practical applications. Several standard bodies, companies, and ad-hoc consortia are proposing their own solution for inter-device communication. In this context, DYnamic, Adaptive MAnagement of Networks and Devices (DYAMAND) was presented in a previous publication to solve the interoperability issues introduced by the multitude of available technologies. In this paper a DYAMAND case study is presented: in cooperation with a large company, a monitoring application was developed for flexible office spaces in order to reliably reorganize an office environment and give real-time feedback on the usage of meeting rooms. Three wireless sensor technologies were investigated to be used in the pilot. The solution was deployed in a "friendly user" setting at a research institute (iMinds) prior to deployment at the large company's premises. Based on the findings of both installations, requirements for an application platform supporting development and management of smart (office) applications were listed. DYAMAND was used as the basis of the implementation. Although the local management of networked devices as provided by DYAMAND enables easier development of intelligent applications, a number of remote services discussed in this paper are needed to enable reliable and up-to-date support (of new technologies).

The Flemish project Linear was a large scale residential demand response pilot that aims to validate innovative smart grid technology building on the rollout of information and communication technologies in the power grid. For this pilot a scalable, reliable and interoperable ICT infrastructure was set up, interconnecting 240 residential power grid customers with the backend systems of energy service providers (ESPs), flexibility aggregators, distribution system operators (DSOs) and balancing responsible parties (BRPs). On top of this architecture several business cases were rolled out, which require the sharing of metering data and flexibility information, and demand response algorithms for the balancing of renewable energy and the mitigation of voltage and power issues in distribution grids. The goal of the pilot is the assessment of the technical and economical feasibility of residential demand response in real life, and of the interaction with the end-consumer. In this paper we focus on the practical experiences and lessons learnt during the deployment of the ICT technology for the pilot. This includes the real-time gathering of measurement data and real-time control of a wide range of smart appliances in the homes of the participants. We identified a number of critical issues that need to be addressed for a future full-scale roll-out: (i) reliable in-house communication, (ii) interoperability of appliances, measurement equipment, backend systems, and business cases, and (iii) sufficient backend processing power for real-time analysis and control.

Optical networks are crucial to support increasingly demanding cloud services. Delivering the requested quality of services (in particular latency) is key to successfully provisioning end-to-end services in clouds. Therefore, as for traditional optical network services, it is of utter importance to guarantee that clouds are resilient to any failure of either network infrastructure (links and/or nodes) or data centers. A crucial concept in establishing cloud services is that of network virtualization: the physical infrastructure is logically partitioned in separate virtual networks. To guarantee end-to-end resilience for cloud services in such a set-up, we need to simultaneously route the services and map the virtual network, in such a way that an alternate routing in case of physical resource failures is always available. Note that combined control of the network and data center resources is exploited, and the anycast routing concept applies: we can choose the data center to provide server resources requested by the customer to optimize resource usage and/or resiliency. This paper investigates the design of scalable optimization models to perform the virtual network mapping resiliently. We compare various resilience options, and analyze their compromise between bandwidth requirements and resiliency quality.

To express a more nuanced notion of relevance as compared to binary judgments, graded relevance levels can be used for the evaluation of search results. Especially in Web search, users strongly prefer top results over less relevant results, and yet they often disagree on which are the top results for a given information need. Whereas previous works have generally considered disagreement as a negative effect, this paper proposes a method to exploit this user disagreement by integrating it into the evaluation procedure.

We address the dimensioning of infrastructure, comprising both network and server resources, for large-scale decentralized distributed systems such as grids or clouds. We design the resulting grid/cloud to be resilient against network link or server failures. To this end, we exploit relocation: Under failure conditions, a grid job or cloud virtual machine may be served at an alternate destination (i.e., different from the one under failure-free conditions). We thus consider grid/cloud requests to have a known origin, but assume a degree of freedom as to where they end up being served, which is the case for grid applications of the bag-of-tasks (BoT) type or hosted virtual machines in the cloud case. We present a generic methodology based on integer linear programming (ILP) that: chooses a given number of sites in a given network topology where to install server infrastructure; and determines the amount of both network and server capacity to cater for both the failure-free scenario and failures of links or nodes. For the latter, we consider either failure-independent (FID) or failure-dependent (FD) recovery. Case studies on European-scale networks show that relocation allows considerable reduction of the total amount of network and server resources, especially in sparse topologies and for higher numbers of server sites. Adopting a failure-dependent backup routing strategy does lead to lower resource dimensions, but only when we adopt relocation (especially for a high number of server sites): Without exploiting relocation, potential savings of FD versus FID are not meaningful.

In this paper we address the problem of dimensioning infrastructure, comprising both network and server resources, for large-scale decentralized distributed systems such as grids or clouds. We will provide an overview of our work in this area, and in particular focus on how to design the resulting grid/cloud to be resilient against network link and/or server site failures. To this end, we will exploit relocation: under failure conditions, a request may be sent to an alternate destination than the one under failure-free conditions. We will provide a comprehensive overview of related work in this area, and focus in some detail on our own most recent work. The latter comprises a case study where traffic has a known origin, but we assume a degree of freedom as to where its end up being processed, which is typically the case for e.g., grid applications of the bag-of-tasks (BoT) type or for providing cloud services. In particular, we will provide in this paper a new integer linear programming (ILP) formulation to solve the resilient grid/cloud dimensioning problem using failure-dependent backup routes. Our algorithm will simultaneously decide on server and network capacity. We find that in the anycast routing problem we address, the benefit of using failure-dependent (FD) rerouting is limited compared to failure-independent (FID) backup routing. We confirm our earlier findings in terms of network capacity savings achieved by relocation compared to not exploiting relocation (order of 6-10% in the current case studies).

Consumer devices increasingly are “smart” and hence offer services that can interwork with and/or be controlled by others. However, the full exploitation of the inherent opportunities this offers, is hurdled by a number of potential limitations. First of all, the interface towards the device might be vendor and even device specific, implying that extra effort is needed to support a specific device. Standardization efforts try to avoid this problem, but within a certain standard ecosystem the level of interoperability can vary (i.e. devices carrying the same standard logo are not necessarily interoperable). Secondly, different application domains (e.g. multimedia vs. energy management) today have their own standards, thus limiting trans-sector innovation because of the additional effort required to integrate devices from traditionally different domains into novel applications. In this paper, we discuss the basic components of current so-called service discovery protocols (SDPs) and present our DYAMAND (DYnamic, Adaptive MAnagement of Networks and Devices) framework. We position this framework as a middleware layer between applications and discoverable/controllable devices, and hence aim to provide the necessary tool to overcome the (intra- and inter-domain) interoperability gaps previously sketched. Thus, we believe it can act as a catalyst enabling transsector innovation.

Distributed renewable power generators, such as solar cells and wind turbines are difficult to predict, making the demand-supply problem more complex than in the traditional energy production scenario. They also introduce bidirectional energy flows in the low-voltage power grid, possibly causing voltage violations and grid instabilities. In this article we describe a distributed algorithm for residential energy management in smart power grids. This algorithm consists of a market-oriented multi-agent system using virtual energy prices, levels of renewable energy in the real-time production mix, and historical price information, to achieve a shifting of loads to periods with a high production of renewable energy. Evaluations in our smart grid simulator for three scenarios show that the designed algorithm is capable of improving the self consumption of renewable energy in a residential area and reducing the average and peak loads for externally supplied power.

The increase of distributed renewable electricity generators, such as solar cells and wind turbines, requires new energy management systems where real-time measurements and communication between end users, suppliers and utilities are vital. To address this need, we propose a common service architecture that allows houses with renewable energy generation and smart energy devices to plug into a distributed energy management system, integrated with the public power grid. The presented architecture facilitates end-users to optimize their energy consumption, enables power network operators to better balance supply and demand, and creates a platform where new market players (e.g. ESCOs) can easily provide new services. This service architecture has been implemented and is currently evaluated in a field trial with 21 users, of which we present the initial results.

The evolution toward grid and cloud computing as observed for over a decennium illustrates the crucial role played by (optical) networks in supporting today's applications. In this paper, we start from an overview of the challenging applications in both academic (further referred to as scientific), enterprise (business) and nonprofessional user (consumer) domains. They pose novel challenges, calling for efficient interworking of IT resources, for both processing and storage, as well as the network that interconnects them and provides access to their users. We outline those novel applications' requirements, including sheer performance attributes (which will determine the quality as perceived by end users of the cloud applications), as well as the ability to adapt to changing demands (usually referred to as elasticity) and possible failures (i.e., resilience). In outlining the foundational concepts that provide the building blocks for grid/cloud solutions that meet the stringent application requirements we highlight, a prominent role is played by optical networking. The pieces of the solution studied in this respect span the optical transport layer as well as mechanisms located in higher layers (e.g., anycast routing, virtualization) and their interworking (e.g., through appropriate control plane extensions and middleware). Based on this study, we conclude by identifying challenges and research opportunities that can enable future-proof optical cloud systems (e.g., pushing the virtualization paradigms to optical networks).

Grids can efficiently deal with challenging computational and data processing tasks which cutting edge science is generating today. So-called e-Science grids cope with these complex task by deploying geographically distributed server infrastructure, interconnected by high speed networks. The latter benefit from optical technology, offering low latencies and high bandwidths, thus giving rise to so-called optical grids or lambda grids. In this paper, we address the dimensioning problem of such grids: how to decide how much server infrastructure to deploy, at which locations in a given topology, the amount of network capacity to provide and which routes to follow along them. Compared to earlier work, we propose an integrated solution solving these questions in an integrated way, i.e., we jointly optimize network and server capacity, and incorporate resiliency against both network and server failures. Assuming we are given the amount of resource reservation requests arriving at each network node (where a resource reservation implies to reserve both processing capacity at a server site, and a network connection towards it), we solve the problem of first choosing a predetermined number of server locations to use, and subsequently determine the routes to follow while minimizing resource requirements. In a case study on a meshed European network comprising 28 nodes and 41 links, we show that compared to classical (i.e. without relocation) shared path protection against link failures only, we can offer resilience against both single link and network failures by adding about 55% extra server capacity, and 26% extra wavelengths.

This paper is aiming at facilitating the energy-efficient operation of an integrated optical network and IT infrastructure. In this context we propose an energy-efficient routing algorithm for provisioning of IT services that originate from specific source sites and which need to be executed by suitable IT resources (e.g. data centers). The routing approach followed is anycast, since the requirement for the IT services is the delivery of results, while the exact location of the execution of the job can be freely chosen. In this scenario, energy efficiency is achieved by identifying the least energy consuming IT and network resources required to support the services, enabling the switching off of any unused network and IT resources. Our results show significant energy savings that can reach up to 55% compared to energy-unaware schemes, depending on the granularity with which a data center is able to switch on/off servers.

People's multimedia content is spread around their home network and content services on the Internet, such as YouTube, Flickr, Facebook. In this paper we present a system that aggregates all the multimedia content of the end user and integrates it into a unified collection for the user's convenience. The system provides location transparency of multimedia content, content filtering on player compatibility and metadata completion to aid in improved usability. This effectively enables the user to rediscover his multimedia collection without any technical knowledge. A proof-of-concept implementation known as Intelligent Distributed Multimedia Collection (IDMC) has been made that is able to detect and browse UPnP MediaServer devices as well as collect information from YouTube. This implementation also contains a media player and is able to control UPnP MediaRenderer devices remotely. Furthermore, performance has been measured to assess different ways of iterating through a multimedia collection.

Nowadays, the Quality of Service (QoS) in Optical Grids has become a key issue. An important QoS factor is resiliency, namely the ability to survive from certain network failures. Although several traditional network protection schemes were devised in the past, they are not optimized for Optical Grids. In an earlier work, we proposed relocation strategies providing backup paths to alternate destinations, exploiting the anycast routing principle of grids. To show the advantage of relocation (compared to traditional network protection) in terms of reduced network capacity, we formulated the network dimensioning problem as an Integer Linear Program (ILP). Yet, its solution exhibited very poor scalability and appeared not practical for reasonably large scale case studies. Therefore, we propose a novel formulation for the relocation protection scheme, using column generation (CG). This approach decomposes the original ILP into two parts, specifically a Restricted Master Problem (RMP) and a Pricing Problem (PP) which are iteratively and alternatively solved until the optimality condition is satisfied. Such a CG decomposition has a significant impact on the complexity of the model, leading to a significant improvement over previous ILPs in term of scalability and running times. We demonstrate that the CG method is highly scalable and generates nearly optimal solutions using case studies with up to 300 connections, showing it to be highly competitive with a previously proposed heuristic. We also perform some comparisons of the anycast scheme with the classical shared path protection on larger network instances.

In order for service providers to provide their users high quality services in the home network, Quality of Service (QoS) provisioning is needed to protect premium services. In this paper, we describe how a Universal Plug-and-Play (UPnP) based home network architecture solves this problem in a heterogeneous home network. We outline how it both relieves the end user from troublesome configuration and still offers control to the service provider. We particularly present performance assessment results for UPnP-QoS v3, based on a fully operational experimental implementation. The quantitative measurement results are further used in extensive simulations demonstrating acceptable response times and clear QoS admission differentiation.

Nowadays, users consume a lot of functionality in their home coming from a service provider located in the Internet. While the home network is typically shielded off as much as possible from the `outside world', the supplied services could be greatly extended if it was possible to use local information. In this article, an extended service is presented that integrates the user's multimedia content, scattered over multiple devices in the home network, into the Electronic Program Guide (EPG) of the Digital TV. We propose to virtualize the set-top box, by migrating all functionality except user interfacing to the service provider infrastructure. The media in the home network is discovered through standard Universal Plug and Play (UPnP), of which the QoS functionality is exploited to ensure high quality playback over the home network, that basically is out of the control of the service provider. The performance of the subsystems are analysed.

Deflection routing is a much-studied contention resolution technique in the context of Optical Burst/Packet Switching networks, as it promises to improve burst blocking performance and may reduce or even eliminate buffer requirements. An OBS-based Grid is frequently based on anycast routing, which h

This paper presents a programmable multi-granular optical cross connect (MG-OXC) and network architecture deployable in multi-service and multi-provider networks. The concept of programmable MG-OXC is introduced to provide a way of utilizing multiple switching/transport granularities to efficiently

To cope with ever-increasing traffic demands in transport networks, all-optical switching is currently perceived as a potential solution to remove bottlenecks caused by optoelectronic conversions. An effective realization of this concept must support a wide range of traffic patterns, while remaining feasible to construct and deploy both in an economical and practical sense. In this paper, we propose the use of multi-granular optical cross-connects (MG-OXC), which support switching on both the wavelength and sub-wavelength level. To this end, the MG-OXCs are equipped with cheap, highly scalable slow switching fabrics, as well as a small number of expensive fast switching ports. The main goal of this work is to motivate the use of multi-granular switching, as this can reduce total network installation costs. To this end, we introduce an Integer Linear Programming model, and our evaluation demonstrates that multi-granular optical switching can be a cost-effective solution on the network level, in comparison to slow only or fast only approaches. Furthermore, we can achieve reduced costs of individual OXC nodes, which allows us to minimize scalability problems corresponding to emerging fast switching fabrics.

Grid computing aims to realize a high-performance computing environment, while increasing the usage efficiency of installed resources. This puts considerable constraints on the network technology, and ultimately has led to the development of Grids over optical networks. In this paper, we investigate the fundamental question of how to optimize the performance of such Grid networks. We start with an analysis of different architectural approaches (and their respective technological choices) to integrate Grid computing with optical networks. This results in models and algorithms to design optical Grid networks, and we show the importance to combine both dimensioning (offline) and scheduling (online) in the design phase of such systems. Finally, the concept of anycast routing is introduced and motivated. Both exact and heuristic algorithms are proposed, and their performance in terms of blocking probability and latency is presented.

Notwithstanding IP anycast's introduction in Internet standards dates back to 1993 and its more recent adoption in IPv6 standards, its use in production environments is limited to date. This is mainly because native IP anycast lacks routing scalability and does not support session-based communications, thereby limiting its applicability to single request-response services such as DNS. For this reason, we propose a transparent anycast overlay architecture that retains the strengths of native anycast and neutralizes above-mentioned limitations. The resulting proxy infrastructure unleashes the power of anycast by opening up new opportunities for transparent distributed service provisioning. Taking into account user demands, available resources, network overhead and anycast infrastructure costs, we provide near- optimal heuristics for the placement of proxy nodes and dimensioning the infrastructure in large networks. We show that even modest overlay infrastructures, consisting of a small number of proxy routers, provide an effective stateful anycast solution where the detour via the proxy routers is negligible in terms of extra network load. Furthermore, simulation results illustrate that server state aggregation in the proxy nodes lessens control plane overhead, which contributes significantly to service robustness.

Optical packet/burst switching is considered a promising technique to improve the performance of optical networks. Key components in these technologies are the optical switching nodes. Some of these node architectures suffer from internal blocking. Synchronous operation allows overcoming most of the problems introduced by this internal blocking. However, in asynchronous networks internal blocking can have a more pronounced effect. In this paper, we propose a windowing technique to improve the performance of internally blocking optical switching nodes in asynchronous operation. Simulations will show significant improvements can be made.

In this paper, promising technologies and a network architecture are presented for future optical packet switched networks. The overall network concept is presented and the major choices are highlighted and compared with alternative solutions. Both long and shorter term approaches are considered, as well as both the wide-area network and multiple-area networks parts of the network. The results presented in this paper were developed in the frame of the research project DAVID (Data And Voice Integration over DWDM) project, funded by the European Commission through the IST-framework.

The explosive growth of data traffic-for example, due to the popularity of the Internet-poses important emerging network requirements on today's telecommunication networks. This paper describes how core networks will evolve to optical transport networks (OTNs), which are optimized for the transport of data traffic, resulting in an IP-directly-over-OTN paradigm. Special attention is paid to the survivability of such data-centric optical networks. This becomes increasingly crucial since more and more traffic is multiplexed onto a single fiber (e.g., 160/spl times/10 Gb/s), implying that a single cable cut can affect incredible large traffic volumes. In particular, this paper is tackling multilayer survivability problems, since a data-centric optical network consists of at least an IP and optical layer. In practice, this means that the questions "in which layer or layers should survivability be provided?" and "if multiple layers are chosen for this purpose, then how should this functionality in these layers be coordinated?" have to be answered. In addition to a theoretical study, some case studies are presented in order to illustrate the relevance of the described issues and to help in strategic planning decisions. Two case studies are studying the problem from a capacity viewpoint. Another case study presents simulations from a timing/throughput performance viewpoint.