Sharing databases and data streams imposes the danger of revealing private information in the form of complex events which can comprise individual data elements and their combinations. Identifying these privacy-revealing complex events is crucial for preserving privacy while maintaining data utility. However, data producers often lack the expertise to comprehensively identify these events, which undermines many state-of-the-art privacy-preserving mechanisms that rely on accurate event labeling. To address this challenge, we developed pArborist - a tool that can semi-automatically create a set of queries to identify and label privacy-revealing complex events in both static datasets and dynamic data streams, guided by the privacy requirements of the data producer. pArborist uses the schema of the database or data stream combined with initial input from the data producer, i.e., seed queries. From each seed query, pArborist grows a tree containing all possible syntactically correct queries, constrained by an upper limit on computational resources. Following this growing phase, the tree is refined by eliminating queries that lack correlation to the seed or are conditionally independent of the seed. Our evaluation indicates that pArborist achieves overall recall of \(90\%\) and precision of \(93\%\) in finding privacy-revealing queries, and this significantly surpasses the state-of-the-art approach FQID. In data stream processing experiments, pArborist introduces a delay of approximately 1.3 ms following an average warm-up period of 920 ms. The experiments also show that pArborist can automatically detect privacy-revealing complex events according to GDPR.

This paper presents a novel approach to requirements engineering for responsible explainable AI (rexAI) that pays attention to the broad range of explanation aspects and enforces the consideration of computer science, ethical, legal, and medical requirements. The medical use-case we are investigating is pediatric sleep apnea diagnosis. We analyze the non-functional and functional requirements and explore the multiple relationships and dependencies between the rexAI requirements and principles.

Good training data is a prerequisite to develop useful Machine Learning applications. However, in many domains existing data sets cannot be shared due to privacy regulations (e.g., from medical studies). This work investigates a simple yet unconventional approach for anonymized data synthesis to enable third parties to benefit from such anonymized data. We explore the feasibility of learning implicitly from visually unrealistic, task-relevant stimuli, which are synthesized by exciting the neurons of a trained deep neural network. As such, neuronal excitation can be used to generate synthetic stimuli. The stimuli data is used to train new classification models. Furthermore, we extend this framework to inhibit representations that are associated with specific individuals. Extensive comparative empirical investigation shows that different algorithms trained on the stimuli are able to generalize successfully on the same task as the original model.

Sleep apnea is a common yet severely under-diagnosed sleep related disorder. Unattended sleep monitoring at home with low-cost sensors can be leveraged for condition detection, and Machine Learning offers a generalized solution for this task. However, patient characteristics, lack of sufficient training data, and other factors can imply a domain shift between training and end-user data and reduced task performance. In this work, we address this issue with the aim to achieve personalization based on the patient’s needs. We present an unsupervised domain adaptation (UDA) solution with the constraint that labeled source data are not directly available. Instead, a classifier trained on the source data is provided. Our solution iteratively labels target data sub-regions based on classifier beliefs, and trains new classifiers from the expanding dataset. Experiments with sleep monitoring datasets and various sensors show that our solution outperforms the classifier trained on the source domain, with a kappa coefficient improvement from 0.012 to 0.242. Additionally, we apply our solution to digit classification DA between three well-established datasets, to investigate its generalizability, and allow for related work comparisons. Even without direct access to the source data, it outperforms several well-established UDA methods in these datasets.

Sleep apnea is a common and strongly under-diagnosed severe sleep-related respiratory disorder with periods of disrupted or reduced breathing during sleep. To diagnose sleep apnea, sleep data are collected with either polysomnography or polygraphy and scored by a sleep expert. We investigate in this work the use of supervised machine learning to automate the analysis of polygraphy data from the A3 study containing more than 7,400 hours of sleep monitoring data from 579 patients. We conduct a systematic comparative study of classification performance and resource use with different combinations of 27 classifiers and four sleep signals. The classifiers achieve up to 0.8941 accuracy (kappa: 0.7877) when using all four signal types simultaneously and up to 0.8543 accuracy (kappa: 0.7080) with only one signal, i.e., oxygen saturation. Methods based on deep learning outperform other methods by a large margin. All deep learning methods achieve nearly the same maximum classification performance even when they have very different architectures and sizes. When jointly accounting for classification performance, resource consumption and the ability to achieve with less training data high classification performance, we find that convolutional neural networks substantially outperform the other classifiers.

Good training data is a prerequisite to develop useful Machine Learning applications. However, in many domains existing data sets cannot be shared due to privacy regulations (e.g., from medical studies). This work investigates a simple yet unconventional approach for anonymized data synthesis to enable third parties to benefit from such anonymized data. We explore the feasibility of learning implicitly from visually unrealistic, task-relevant stimuli, which are synthesized by exciting the neurons of a trained deep neural network. As such, neuronal excitation can be used to generate synthetic stimuli. The stimuli data is used to train new classification models. Furthermore, we extend this framework to inhibit representations that are associated with specific individuals. We use sleep monitoring data from both an open and a large closed clinical study, and Electroencephalogram sleep stage classification data, to evaluate whether (1) end-users can create and successfully use customized classification models, and (2) the identity of participants in the study is protected. Extensive comparative empirical investigation shows that different algorithms trained on the stimuli are able to generalize successfully on the same task as the original model. Architectural and algorithmic similarity between new and original models play an important role in performance. For similar architectures, the performance is close to that of using the original data (e.g., Accuracy difference of 0.56%-3.82%, Kappa coefficient difference of 0.02-0.08). Further experiments show that the stimuli can provide state-ofthe-art resilience against adversarial association and membership inference attacks.

Supervised machine learning applications in the health domain often face the problem of insufficient training datasets. The quantity of labelled data is small due to privacy concerns and the cost of data acquisition and labelling by a medical expert. Furthermore, it is quite common that collected data are unbalanced and getting enough data to personalize models for individuals is very expensive or even infeasible. This paper addresses these problems by (1) designing a recurrent Generative Adversarial Network to generate realistic synthetic data and to augment the original dataset, (2) enabling the generation of balanced datasets based on heavily unbalanced dataset, and (3) to control the data generation in such a way that the generated data resembles data from specific individuals. We apply these solutions for sleep apnea detection and study in the evaluation the performance of four well-known techniques, i.e., K-Nearest Neighbour, Random Forest, Multi-Layer Perceptron, and Support Vector Machine. All classifiers exhibit in the experiments a consistent increase in sensitivity and a kappa statistic increase by between 0.007 and 0.182.

As a step towards mobile and ubiquitous computing, the TRAMP Real-time Application Mobility Platform (TRAMP) enables secure and user-friendly migration of multimedia application components. The motivating factors are device and user mobility, i.e., to let the user determine where application components should execute at all times. In order to organize and locate roaming devices, a decentralized peer-to-peer overlay is developed. To provide location transparency for application developers and enable efficient data sharing among application components, an API is provided that hides whether shared memory or latency optimized network distribution is used to share data. The problem of open TCP connections to third-party applications is solved with a proxy-based approach. Multimedia applications are especially sensitive to service interruptions that can be caused by migration, and a new approach for check-pointing, state transfer, and resuming an application with minimal freeze time is therefore introduced. The evaluation of TRAMP shows that this optimization for low freeze time has significant effect. For example, migrating the Skype video-conferencing application with TRAMP reduces the average freeze time from 41.1 seconds with legacy approaches to 213 milliseconds. Further results from the empirical evaluation of TRAMP demonstrate that it has no negative impact on the Quality of Experience.

Network simulation is commonly used to evaluate the performance of distributed systems, but these approaches do not account for the performance impact that protocol execution on nodes has on performance, which may be significant. We propose a methodology to capture execution models from communication software running on real devices where the execution models can be integrated with discrete event network simulators to improve their accuracy. We provide a set of rules to instrument the software to obtain the events of importance, and present techniques to create executable models based on the obtained traces. To make the models scalable, processing stages are reduced to statistical distributions. When the resulting models are executed in a device model with a scheduler simulator, we are able to model the dynamics of multithreading and parallel execution. Our initial results from a proof-of-concept extension to Ns-3 show that our models are able to accurately model protocol execution on the Google Nexus One with low simulation overhead.

Due to the lack of multicast services in the Internet, applications based on single-source, multiple-destination transfers such as video conferencing, IP radio, and IPTV must use unicast. This type of traffic exhibits high redundancy with temporal clustering of duplicated packets. The redundancy originates from multiple transfers of the same data chunk over the same link. We propose CacheCast, a link-layer caching mechanism that eliminates the redundant data transmissions using small caches on links. CacheCast's underlying principles are simplicity and reliability. It is a fully distributed and incrementally deployable architecture. It consists of small caches on links that act independently and a server support that simplifies the link cache operation. Our analysis indicates that transfers of the same data to multiple destinations with CacheCast can achieve near-multicast efficiency in terms of consumed link bandwidth. The implementation of CacheCast proves its feasibility, efficiency, and the improvements of the server.

Network simulators typically do not include node models. Our studies show that in networks such as mobile networks, the impact of nodes on performance can be significant. Existing techniques to simulate nodes' are not scalable for network simulations, and require a too large modelling effort to be feasible for network research. In this paper, we propose to capture flexible per-protocol performance profiles from real, running systems using instrumentation and traffic benchmarking techniques. By using the obtained profiles as input into an extended scheduler simulator, the behaviour of the node can be accurately reproduced. Since the processing overhead is represented statistically, we preserve scalability and a low modelling overhead.

Due to the lack of multicast services in the Internet, applications based on single source multiple destinations transfers such as video conferencing, IP radio, IPTV must use unicast or application layer multicast. This in turn has several well-known drawbacks. A basic insight is that this type of traffic exhibits high redundancy with temporal clustering of duplicated packets. The redundancy originates from multiple transfers of the same data chunk over the same link. We propose CacheCast-a link layer caching mechanism-that eliminates the redundant data transmissions using small caches on links. CacheCast's underlying principles are simplicity and reliability. It is a fully distributed and incrementally deployable architecture. It consists of small caches on links that act independently. A single cache removes redundant data from a packet on the link entry and recovers the data on the link exit. Thus, link caches are transparent to routers. We show through analysis and simulation that CacheCast achieves near multicast efficiency for superposition of unicast connections. We implemented CacheCast in ns-2 and show that it does not violate the current understanding of "fairness" in the Internet.

Using Mobile Ad-hoc Networks (MANETs) for audio and video transmission is very promising for application domains such as emergency and rescue. However, audio/video streaming services are not designed for such dynamic and unstable networks. The problems are even more important in so-called sparse MANETs where the node density is relatively low so that disconnections and network partitions are common. We have designed an architecture that combines MANET routing with caching and delay tolerant store-carry-forward operations in an overlay network to improve the quality of audio/video transmission over sparse MANETs. We have implemented a prototype to evaluate the architecture. The results from the experiments demonstrate that our system clearly outperforms simple client-server solutions when the network has temporal disconnections.

Today's Internet has evolved from an early experimental small-scale network to a world-wide network which infrastructure that is used on a day to day basis by companies, public bodies, governments, individuals etc. Many shortcomings of the Internet have been addressed over the last years as "add-ons", but recently many "Future Internet" initiatives have been started with the challenge to re-design the Internet to meet the future challenges. Since the majority of multimedia systems and applications are depending on the Internet we aim in this tutorial to give an overview on ongoing initiatives, identify new concepts and study how they might be leveraged for multimedia systems and applications.

The Streaming Peer-to-Peer Protocol (SPP) architecture is a hybrid P2P system for interactive Video-on-Demand Streaming. We aim to demonstrate key issues of SPP, i.e., peer selection and incentive management. Through real live demonstration on PlanetLab and through simulated scenarios of interest in NS2, the audience will be able to judge the resulting perceptual Quality-of-Service (QoS) on a large screen connected to a set-top box. Furthermore, we plan to visualize what is happening inside SPP on a Laptop.

In this paper we discuss how dynamic updates in a sparse mobile ad-hoc network can be handled in a rescue scenario. As rescue operations have an organisational and operational structure, updates can take advantage of rescue operation roles and lines of reporting, together with update priorities. To illustrate this, we give a short description of our rescue scenario, together with an outline of our multi-tier data dictionary solution that makes use of semantic linked distributed data dictionaries and ontology enhanced metadata descriptions. We present an example ontology of profiles for user, device, rescue scenario, and information items, tailored for ontology based dynamic updates.

In this paper, we present APEX, a disk scheduling framework with QoS support, designed for environments with highly varying disk bandwidth usage. In particular, we focus on a Learning-on-Demand scenario supported by a multimedia database management system, where students can search for, and play back multimedia-based learning material. APEX is based on a two-level scheduling architecture, where the upper level realizes different service classes using a set of queues, while the lower level distributes available disk bandwidth among these queues.In this paper, we focus on the low-level scheduling in APEX, which is based on an extended token bucket algorithm. The disk requests scheduled for service are assembled into batches, which render possible good efficiency for the disk. Combined with a very efficient work-conservation scheme, this enables APEX to apply bandwidth where it is needed, without efficiency loss. We demonstrate, through simulations, that APEX provides both higher throughput and lower response times than other mixed-media disk schedulers, while still avoiding deadline violations for real-time requests. We also show its robustness with respect to misaligned bandwidth allocation.

Data servers for multimedia applications like news-on-demand represent a severe bottleneck, because a potentially (very) high number of users concurrently retrieve data with high data rates. In the Intermediate Storage Node Concept (INSTANCE) project, we develop a new architecture for media-on-demand servers that maximizes the number of concurrent clients a single server can support. Traditional bottlenecks, like copy operations, multiple copies of the same data element in main memory, and checksum calculation in communication protocols are avoided by applying three orthogonal techniques: zero-copy-one-copy memory architecture, network level framing, and integrated error management. We describe the design, implementation, and evaluation of our integrated error management mechanism. Our results show that the reuse of parity information from a RAID system as forward error correction information in the transport protocol reduces the server workload and enables smooth playout at the client.

Article Techniques to increase disk access locality in the Minorca multimedia file system Share on Authors: Chuanbao Wang University of Oslo, UniK - Center for Technology at Kjeller Granaveien 33, P.B. 70 N-2027 Kjeller, Norway University of Oslo, UniK - Center for Technology at Kjeller Granaveien 33, P.B. 70 N-2027 Kjeller, NorwayView Profile , Vera Goebel University of Oslo, UniK - Center for Technology at Kjeller Granaveien 33, P.B. 70 N-2027 Kjeller, Norway University of Oslo, UniK - Center for Technology at Kjeller Granaveien 33, P.B. 70 N-2027 Kjeller, NorwayView Profile , Thomas Plagemann University of Oslo, UniK - Center for Technology at Kjeller Granaveien 33, P.B. 70 N-2027 Kjeller, Norway University of Oslo, UniK - Center for Technology at Kjeller Granaveien 33, P.B. 70 N-2027 Kjeller, NorwayView Profile Authors Info & Claims MULTIMEDIA '99: Proceedings of the seventh ACM international conference on Multimedia (Part 2)October 1999 Pages 147–150https://doi.org/10.1145/319878.319917Online:01 October 1999Publication History 6citation202DownloadsMetricsTotal Citations6Total Downloads202Last 12 Months1Last 6 weeks1 Get Citation AlertsNew Citation Alert added!This alert has been successfully added and will be sent to:You will be notified whenever a record that you have chosen has been cited.To manage your alert preferences, click on the button below.Manage my AlertsNew Citation Alert!Please log in to your account Save to BinderSave to BinderCreate a New BinderNameCancelCreateExport CitationPublisher SiteGet Access