Search

The detection of events in time series is an important task in several areas of knowledge where operations monitoring is essential. Experts often have to deal with choosing the most appropriate event detection method for a time series, which can be a complex task. There is a demand for benchmarking different methods in order to guide this choice. For this, standard classification accuracy metrics are usually adopted. However, they are insufficient for a qualitative analysis of the tendency of a method to precede or delay event detections. Such analysis is interesting for applications in which tolerance for "close" detections is important rather than focusing only on accurate ones. In this context, this paper proposes a more comprehensive event detection benchmark process, including an analysis of temporal bias of detection methods. For that, metrics based on the time distance between event detections and identified events (detection delay) are adopted. Computational experiments were conducted using real-world and synthetic datasets from Yahoo Labs and resources from the Harbinger framework for event detection. Adopting the proposed detection delay-based metrics helped obtain a complete overview of the performance and general behavior of detection methods.

Consider a set of black-box models – each of them independently trained on a different dataset – answering the same predictive spatio-temporal query. Being built in isolation, each model traverses its own life-cycle until it is deployed to production, learning data patterns from different datasets and facing independent hyper-parameter tuning. In order to answer the query, the set of black-box predictors has to be ensembled and allocated to the spatio-temporal query region. However, computing an optimal ensemble is a complex task that involves selecting the appropriate models and defining an effective allocation strategy that maps the models to the query region. In this paper we present DJEnsemble, a cost-based strategy for the automatic selection and allocation of a disjoint ensemble of black-box predictors to answer predictive spatio-temporal queries. We conduct a set of extensive experiments that evaluate DJEnsemble and highlight its efficiency, selecting model ensembles that are almost as efficient as the optimal solution. When compared against the traditional ensemble approach, DJEnsemble achieves up to 4X improvement in execution time and almost 9X improvement in prediction accuracy.

Predicting the physical or functional associations through protein-protein interactions (PPIs) represents an integral approach for inferring novel protein functions and discovering new drug targets during repositioning analysis. Recent advances in high-throughput data generation and multi-omics techniques have enabled large-scale PPI predictions, thus promoting several computational methods based on different levels of biological evidence. However, integrating multiple results and strategies to optimize, extract interaction features automatically and scale up the entire PPI prediction process is still challenging. Most procedures do not offer an in-silico validation process to evaluate the predicted PPIs. In this context, this paper presents the PredPrIn scientific workflow that enables PPI prediction based on multiple lines of evidence, including the structure, sequence, and functional annotation categories, by combining boosting and stacking machine learning techniques. We also present a pipeline (PPIVPro) for the validation process based on cellular co-localization filtering and a focused search of PPI evidence on scientific publications. Thus, our combined approach provides means to extensive scale training or prediction of new PPIs and a strategy to evaluate the prediction quality. PredPrIn and PPIVPro are publicly available at https://github.com/YasCoMa/predprin and https://github.com/YasCoMa/ppi_validation_process.

In machine learning, Reinforcement Learning (RL) is an important tool for creating intelligent agents that learn solely through experience. One particular subarea within the RL domain that has received great attention is how to define macro-actions, which are temporal abstractions composed of a sequence of primitive actions. This subarea, loosely called skill acquisition, has been under development for several years and has led to better results in a diversity of RL problems. Among the many skill acquisition approaches, graph-based methods have received considerable attention. This survey presents an overview of graph-based skill acquisition methods for RL. We cover a diversity of these approaches and discuss how they evolved throughout the years. Finally, we also discuss the current challenges and open issues in the area of graph-based skill acquisition for RL.

Recent concepts such as Smart Cities, Urban Computing, and Geographic Information Systems are being discussed in various international forums, using themes such as sustainability and efficient use of the city infrastructures. One important aspect in this regard is to correctly associate computational techniques with statistical models and integrate heterogeneous data sources using open data shared by cities. Based on that, this study uses open data from the city of Curitiba (Brazil) in order to bring results on the spatiotemporal evolution of business activities along a period of over thirty years. To that end, the study identifies and discusses important challenges that had to be tackled toward data quality, data categorization, and data integration, in order to perform this type of study in practice. By looking at the dynamics of geographically grounded microeconomic variables, this study shows how the expansion and diversification of business types in different neighborhoods happened, contributing to a better understanding of the process of evolution of the business activity in a city.

The literature is rich in mobility models that aim at predicting human mobility. Yet, these models typically consider only a single kind of data source, such as data from mobile calls or location data obtained from GPS and web applications. Thus, the robustness and effectiveness of such data-driven models from the literature remain unknown when using heterogeneous types of data. In contrast, this paper proposes a novel family of data-driven models, called MobHet, to predict human mobility using heterogeneous data sources. Our proposal is designed to use a combination of features capturing the popularity of a region, the frequency of transitions between regions, and the contacts of a user, which can be extracted from data obtained from various sources, both separately and conjointly. We evaluate the MobHet models, comparing them among themselves and with two single-source data-driven models, namely SMOOTH and Leap Graph, while considering different scenarios with single as well as multiple data sources. Our experimental results show that our best MobHet model produces results that are better than or at least comparable to the best baseline in all considered scenarios, unlike the previous models whose performance is very dependent on the particular type of data used. Our results thus attest the robustness of our proposed solution to the use of heterogeneous data sources in predicting human mobility.

The Linking Open Data (LOD) cloud is a global data space for publishing and linking structured data on the Web. The idea is to facilitate the integration, exchange, and processing of data. The LOD cloud already includes a lot of datasets that are related to the biological area. Nevertheless, most of the datasets about protein interactions do not use metadata standards. This means that they do not follow the LOD requirements and, consequently, hamper data integration. This problem has impacts on the information retrieval, specially with respect to datasets provenance and reuse in further prediction experiments. This paper proposes an ontology to describe and unite the four main kinds of data in a single prediction experiment environment: (i) information about the experiment itself; (ii) description and reference to the datasets used in an experiment; (iii) information about each protein involved in the candidate pairs. They correspond to the biological information that describes them and normally involves integration with other datasets; and, finally, (iv) information about the prediction scores organized by evidence and the final prediction. Additionally, we also present some case studies that illustrate the relevance of our proposal, by showing how queries can retrieve useful information.

We propose a method for the Distributed Assessment of the Closeness CEntrality Ranking (DACCER) in complex networks. DACCER computes centrality based only on localized information restricted to a limited neighborhood around each node, thus not requiring full knowledge of the network topology. We indicate that the node centrality ranking computed by DACCER is highly correlated with the node ranking based on the traditional closeness centrality, which requires high computational costs and full knowledge of the network topology by the entity responsible for calculating the centrality. This outcome is quite useful given the vast potential applicability of closeness centrality, which is seldom applied to large-scale networks due to its high computational costs. Results indicate that DACCER is simple, yet efficient, in assessing node centrality while allowing a distributed implementation that contributes to its performance. This also contributes to the practical applicability of DACCER to the analysis of large complex networks, as indicated in our experimental evaluation using both synthetically generated networks and real-world network traces of different kinds and scales.

Despite the large variety and wide adoption of different techniques to detect and filter unsolicited messages (spams), the total amount of such messages over the Internet remains very large. Some reports point out that around 80% of all emails are spams. As a consequence, significant amounts of network resources are still wasted as filtering strategies are usually performed only at the email destination server. Moreover, a considerable part of these unsolicited messages is sent by users who are unaware of their spamming activity and may thus inadvertently be classified as spammers. In this case, these oblivious users act as spambots, i.e., members of a spamming botnet. This paper proposes a new method for detecting spammers at the source network, whether they are individual malicious users or oblivious members of a spamming botnet. Our method, called SpaDeS, is based on a supervised classification technique and relies only on network-level metrics, thus not requiring inspection of message content. We evaluate SpaDeS using real datasets collected from a Brazilian broadband ISP. Our results show that our method is quite effective, correctly classifying the vast majority (87%) of the spammers while misclassifying only around 2% of the legitimate users.

We propose a platform for the rapid prototyping of active measurement tools to collect network characteristics. Our proposal, named FLAME (Flexible Lightweight Active Measurement Environment), allows the prototyping of active measurement tools that depend or not on cooperative destination nodes. FLAME provides its users with basic active measurement primitives-corresponding to basic building blocks-upon which sophisticated active measurement tools can be prototyped quickly, practically, and efficiently through scripts in the Lua scripting language. This platform is based on the distributed deployment of measurement agents in some network nodes. These agents can send and receive probes in response to Lua scripts sent by a user console. Such agents also publish the measurement data in a standardized way in a repository that a central manager keeps, thus simplifying the management and analysis of such data. We validate the FLAME platform as well as show its flexibility and accuracy through experiments on a local testbed and also on Planet-lab. We also evaluate the proposed platform as compared to previous work in terms of the easing of development due to prototype scripting as well as the achieved memory footprint. In conclusion, we discuss different possibilities for further developments built upon the basic functionality of the proposed platform.

The performance of parallel query processing in a cluster of index servers is crucial for modern web search systems. In such a scenario, the response time basically depends on the execution time of the slowest server to generate a partial ranked answer. Previous approaches investigate performance issues in this context using simulation, analytical modeling, experimentation, or a combination of them. Nevertheless, these approaches simply assume balanced execution times among homogeneous servers (by uniformly distributing the document collection among them, for instance)-a scenario that we did not observe in our experimentation. On the contrary, we found that even with a balanced distribution of the document collection among index servers, correlations between the frequency of a term in the query log and the size of its corresponding inverted list lead to imbalances in query execution times at these same servers, because these correlations affect disk caching behavior. Further, the relative sizes of the main memory at each server (with regard to disk space usage) and the number of servers participating in the parallel query processing also affect imbalance of local query execution times. These are relevant findings that have not been reported before and that, we understand, are of interest to the research community.

The main purpose of this article is to provide researchers or practitioners with greater flexibility to analyze their proposed designs for network, protocol, or video coding of MDC video transmission without being obliged to consider a particular tool set for each kind of video-quality evaluation. Therefore, a generalized simulation and evaluation frame work for video transmission is an important and useful tool to the video-coding of networking was found by research community. On the one hand, for video-coding studies, one can simulate the effects of a more realistic network on video sequences resulting from their coding schemes. On the other hand, in networking analysis, one can evaluate the impact of real video streams on proposed network protocols

Several previous mobility models aim at describing or predicting human behavior in a particular region during a certain period of time. Nevertheless, most of those models have been evaluated using data from a single source, such as data from mobile calls or GPS data obtained from Web applications. Thus, the effectiveness of such models when using different types of data remains unknown. This paper proposes a new model to predict human mobility, called MobDatU, which was designed to use data from mobile calls and data from georeferenced applications (in an isolated or combined way). MobDatU as well as two state-of-the-art models, namely SMOOTH and Leap Graph, are evaluated considering various scenarios with single data source and multiple data sources. The experiments indicate that MobDatU always produces results that are better than or at least comparable to the best baseline in all scenarios, unlike the previous models whose performance is very dependent on the particular type of data used.

International audience; Location-aware applications take into account from where the users are accessing and thereby can offer novel functionalities in the Internet. This paper focuses on improving the accuracy of a geographic location service that relies on delay measurements to locate Internet hosts. Host locations are inferred by comparing delay patterns of geographically distributed landmarks, which are hosts with a known geographic location, with the delay pattern of the target host to be located. We deal with two problems that influence the accuracy of the resulting location estimation: (i) the placement of the landmarks and the probe machines that perform the delay measurements; and (ii) how to best measure the similarity between the delay patterns of the landmarks and the one observed for the target host. For the landmark placement problem, we propose a demographic approach to improve the representativeness of each landmark with respect to the hosts to be located. Given a limited number of landmarks, results show that a demographic placement provides closer landmarks and more accurate location estimations for most hosts. Concerning the placement of probe machines, we show that they have to be sparsely placed to avoid gathering redundant data. Furthermore, we define and evaluate three similarity models. Experiments show that other similarity models outperform the commonly adopted Euclidean distance, resulting then in a more accurate geographic location of Internet hosts.

We propose a method for the Distributed Assessment of the Closeness CEntrality Ranking (DACCER) in complex networks. DACCER computes centrality based only on localized information restricted to a given neighborhood around each node, thus not requiring full knowledge of the network topology. We show that the node centrality ranking computed by DACCER is highly correlated with the node ranking based on the traditional closeness centrality, which requires high computational costs and full knowledge of the network topology. This outcome is quite useful given the vast potential applicability of closeness centrality, which is seldom applied to large-scale networks due to its high computational costs. Results indicate that DACCER is simple, yet efficient, in assessing node centrality while allowing a distributed implementation that contributes to its performance. This also contributes to the practical applicability of DACCER in the analysis of large-scale complex networks, as we show using in our experimental evaluation both synthetically generated networks and traces of real-world networks of different kinds and scales.

We propose a platform for the rapid prototyping of active measurement tools to collect network characteristics. The proposed platform provides its users with basic active measurement primitives upon which sophisticated active measurement tools can be prototyped quickly, practically, and efficiently through scripts in the Lua scripting language. We validate the platform as well as show its flexibility and accuracy through experiments on a local testbed and also on Planet-lab.

Modern large-scale grid computing systems for processing advanced science and engineering applications rely on geographically distributed clusters. In such highly distributed environments, estimating the available bandwidth between clusters is a key issue for efficient task scheduling. We analyze the performance of two well known available bandwidth estimation tools, pathload and abget, with the aim of using them in grid environments. Differently than previous investigations (Jain et al., http://www.caida.org/workshops/isma/0312/slides/rprasad-best.pdf ; Shriram et al., in Passive and active network measurement: 6th international workshop, PAM 2005. Springer, Berlin, 2005), our experiments consider a series of relevant metrics such as accuracy of the estimation, convergence time, degree of intrusion in the grid links, and ability to handle multiple simultaneous estimations. No previous work has analyzed the use of available bandwidth tools for the derivation of efficient grid scheduling.

Mobile devices are increasingly being incorporated onto Picture Archiving and Communication Systems (PACS). Previous work on medical image access using mobile devices essentially focuses on enabling the visualization of the medical images at the mobile devices. In contrast, we propose and develop a distributed system that allows medical image analysis using mobile devices. As a proof-of-concept, based on our distributed system, we develop a tool for dental implant simulation using mobile devices. We also discuss the perspectives of extending the current system to incorporate Computer Aided Diagnosis (CAD) algorithms in it.

Network managers face nowadays a challenging problem to detect traffic from Skype, a very popular application for VoIP communications. If no restrictive firewalls are adopted, Skype uses UDP as its preferred transport protocol, but it is known that due to its high capacity of adaptation, Skype can operate behind many firewalls and network proxies without user configuration. Behind restrictive firewalls, Skype uses Web TCP ports (80 or 443) as a fallback mechanism to delude firewalls and other network elements. This strategy renders Skype traffic disguised as Web traffic quite difficult to detect by network operators. In this paper, we propose a method to efficiently detect Skype flows hidden among Web traffic. We validate our proposal using real-world experimental data gathered at a commercial Internet service provider (ISP) and an academic institution. Our experimental results show a performance of around 90% detection rate of disguised Skype flows with a false positive rate of only 2%, whereas a 100% detection rate of Skype flows in Web traffic is achieved with a false positive rate limited to only 5%. We also evaluate the feasibility of our proposal in a real-time Skype detection scenario.

Skype is a Voice over IP (VoIP) Internet application that is gaining huge popularity in recent years. A key point to Skype popularity is its capability to dynamically adapt itself to operate behind firewalls or network proxies. A common way adopted by Skype to delude these network devices is to use port 80, normally expected to comprise HTTP traffic. In this paper, we propose metrics and investigate statistical tests intended to clearly distinguish Skype flows from HTTP traffic. We validate our study using real-world experimental datasets gathered at a commercial Internet Service Provider (ISP). Our experimental results suggest that the proposed methodology may be seen as a promising building block towards a system to detect general protocol anomalies in HTTP traffic.

There has been a good amount of research over the past few years on the integration of mobile devices into computational grids. Little of it, however, has addressed the more challenging issue of dynamically establishing spontaneous, multihop ad hoc mobile grids, and then only in preliminary form. In such grids, resource discovery and scheduling have been traditionally approached separately in the literature. In our view, this is an inappropriate dichotomy as regards multihop ad hoc mobile grids. In this paper, we propose a resource discovery and scheduling protocol for mobility (DICHOTOMY), which allows resource provisioning to be scheduled among the most resourceful nodes in the mobile grid, while mitigating the overhead of discovery messages exchanged among them. Our experimental results in a testbed and in a simulation platform show that the protocol provides efficient load balancing between nodes under an increasing number of requests, and scales well with respect to an increasing number of nodes in comparison with traditional broadcast-based solutions for service discovery.

Distributed Denial of Service (DDoS) attacks represent a serious threat to the appropriate operation of Internet services. To deal with this threat, we propose an IP traceback system (IPTS) intended to be deployed at the level of Autonomous Systems (ASes). Our IPTS requires a priori no knowledge of the network topology while allowing single-packet trace-back and incremental deployment.

Grids are likely to be the mainstay of distributed computing. Grid was firstly applied as a concept for sharing computing resources among wired nodes, but there has been a growing interest in extending this concept to mobile environments. Although significant work has been done towards mobile grids, much of it has followed centralized approaches based on infrastructure wireless networks. We believe that a less restrictive, decentralized approach that supports mobile collaboration in ad-hoc wireless networks can cater for novel grid applications. To address this issue, we propose a middleware architecture called MoGrid. MoGrid orchestrates the distribution of grid tasks among mobile devices in a peer-to-peer (P2P) fashion. In this paper, we focus on the P2P Discovery Protocol (P2PDP), which is a central element of our architecture. P2PDP aids in distributing tasks among the most resourceful devices, while mitigating the overhead of control messages exchanged among them. We describe a prototype implementation of our architecture and discuss some issues related to the adoption of P2PDP as an ad-hoc resource discovery mechanism in mobile grids.

Multimedia delivery systems, such as Content Distribution Networks (CDNs), improve by knowing the geographic location of their clients. Therefore, we focus on a measurement-based geographic location service of Internet hosts. Such a service infers host locations from delay measurements taken from probe machines toward landmarks, which are hosts with a known geographic location, and the host to be located. We aim at mitigating the number of measurements generated in the network. We propose a two-tier hierarchical structure of landmarks to geographically locate an Internet host as opposed to a flat structure. In our two-tier structure, the upper level mitigates long distance measurements and the lower level keeps measurements within restricted areas. As a consequence, the two-tier geographic location structure significantly reduces the number of measurements and thereby it favors scalability.

International audience; The MPEG format is becoming attractive for the distribution of audio and video over the Internet because of its wide adoption. Nevertheless, the quality perceived by a user of a video transmission may be strongly degraded by indiscriminate discards of packets under the best-effort model. The Differentiated Services (DS) architecture proposes to overcome the limitations for quality of service (QoS) provisioning of today's best-effort Internet. This paper provides a quantitative study and assessment through simulation on how a DS mechanism can improve the delivery quality of MPEG video streams. We use a drop differentiation mechanism to avoid the error propagation effect of a dropped packet in the hierarchical structure of frames in MPEG systems. Results show that the drop differentiation mechanism improves the delivery quality of MPEG video streams while avoiding the waste of network resources, even under heavy load conditions.

International audience; Geolocation of Internet hosts enables a new class of location-aware applications. Previous measurement-based approaches use reference hosts, called landmarks, with a well-known geographic location to provide the location estimation of a target host. This leads to a discrete space of answers, limiting the number of possible location estimates to the number of adopted landmarks. In contrast, we propose Constraint-Based Geolocation (CBG), which infers the geographic location of Internet hosts using multilateration with distance constraints to establish a continuous space of answers instead of a discrete one. However, to use multilateration in the Internet, the geographic distances from the landmarks to the target host have to be estimated based on delay measurements between these hosts. This is a challenging problem because the relationship between network delay and geographic distance in the Internet is perturbed by many factors, including queueing delays and the absence of great-circle paths between hosts. CBG accurately transforms delay measurements to geographic distance constraints, and then uses multilateration to infer the geolocation of the target host. Our experimental results show that CBG outperforms previous geolocation techniques. Moreover, in contrast to previous approaches, our method is able to assign a confidence region to each given location estimate. This allows a location-aware application to assess whether the location estimate is sufficiently accurate for its needs.

International audience; A whole new class of location-aware services may be envisaged by the deployment of a geographic location service for Internet hosts. We focus on a technique that relies on delay measurements and the exploitation of a possible correlation between geographic distance and network delay. Our investigation shows that such a correlation becomes stronger as connectivity becomes richer. Host locations are inferred by comparing delay patterns of geographically distributed landmarks (hosts with a known geographic location) with the delay pattern of the target host to be located. The problem we deal with is how to best measure the similarity between the delay patterns of the landmarks and the one observed for the target host. The location estimation of the target host is the location of the landmark that presents the most similar delay pattern with respect to the one of the target host. We define and evaluate three similarity models. Experiments show that the previously adopted Euclidean distance is outperformed by other similarity models, resulting in a more accurate Internet host location from delay measurements.