Be Sparse Be Dense Be Robust
DOWNLOAD
Download Be Sparse Be Dense Be Robust PDF/ePub or read online books in Mobi eBooks. Click Download or Read Online button to get Be Sparse Be Dense Be Robust book now. This website allows unlimited access to, at the time of writing, more than 1.5 million titles, including hundreds of thousands of titles in various foreign languages. If the content not found or just blank you must refresh this page
Be Sparse Be Dense Be Robust
DOWNLOAD
Author : Sorge, Manuel
language : en
Publisher: Universitätsverlag der TU Berlin
Release Date : 2017-05-31
Be Sparse Be Dense Be Robust written by Sorge, Manuel and has been published by Universitätsverlag der TU Berlin this book supported file pdf, txt, epub, kindle and other format this book has been release on 2017-05-31 with Mathematics categories.
In this thesis we study the computational complexity of five NP-hard graph problems. It is widely accepted that, in general, NP-hard problems cannot be solved efficiently, that is, in polynomial time, due to many unsuccessful attempts to prove the contrary. Hence, we aim to identify properties of the inputs other than their length, that make the problem tractable or intractable. We measure these properties via parameters, mappings that assign to each input a nonnegative integer. For a given parameter k, we then attempt to design fixed-parameter algorithms, algorithms that on input q have running time upper bounded by f(k(q)) * |q|^c , where f is a preferably slowly growing function, |q| is the length of q, and c is a constant, preferably small. In each of the graph problems treated in this thesis, our input represents the setting in which we shall find a solution graph. In addition, the solution graphs shall have a certain property specific to our five graph problems. This property comes in three flavors. First, we look for a graph that shall be sparse! That is, it shall contain few edges. Second, we look for a graph that shall be dense! That is, it shall contain many edges. Third, we look for a graph that shall be robust! That is, it shall remain a good solution, even when it suffers several small modifications. Be sparse! In this part of the thesis, we analyze two similar problems. The input for both of them is a hypergraph H , which consists of a vertex set V and a family E of subsets of V , called hyperedges. The task is to find a support for H , a graph G such that for each hyperedge W in E we have that G[W ] is connected. Motivated by applications in network design, we study SUBSET INTERCONNECTION DESIGN, where we additionally get an integer f , and the support shall contain at most |V| - f + 1 edges. We show that SUBSET INTERCONNECTION DESIGN admits a fixed-parameter algorithm with respect to the number of hyperedges in the input hypergraph, and a fixed-parameter algorithm with respect to f + d , where d is the size of a largest hyperedge. Motivated by an application in hypergraph visualization, we study r-OUTERPLANAR SUPPORT where the support for H shall be r -outerplanar, that is, admit a edge-crossing free embedding in the plane with at most r layers. We show that r-OUTER-PLANAR SUPPORT admits a fixed-parameter algorithm with respect to m + r , where m is the number of hyperedges in the input hypergraph H. Be dense! In this part of the thesis, we study two problems motivated by community detection in social networks. Herein, the input is a graph G and an integer k. We look for a subgraph G' of G containing (exactly) k vertices which adheres to one of two mathematically precise definitions of being dense. In mu-CLIQUE, 0 < mu <= 1, the sought k-vertex subgraph G' should contain at least mu time k choose 2 edges. We study the complexity of mu-CLIQUE with respect to three parameters of the input graph G: the maximum vertex degree delta, h-index h, and degeneracy d. We have delta >= h >= d in every graph and h as well as d assume small values in graphs derived from social networks. For delta and for h, respectively, we obtain fixed-parameter algorithms for mu-CLIQUE and we show that for d + k a fixed-parameter algorithm is unlikely to exist. We prove the positive algorithmic results via developing a general framework for optimizing objective functions over k-vertex subgraphs. In HIGHLY CONNECTED SUBGRAPH we look for a k-vertex subgraph G' in which each vertex shall have degree at least floor(k/2)+1. We analyze a part of the so-called parameter ecology for HIGHLY CONNECTED SUBGRAPH, that is, we navigate the space of possible parameters in a quest to find a reasonable trade-off between small parameter values in practice and efficient running time guarantees. The highlights are that no 2^o(n) * n^c -time algorithms are possible for n-vertex input graphs unless the Exponential Time Hypothesis fails; that there is a O(4^g * n^2)-time algorithm for the number g of edges outgoing from the solution G; and we derive a 2^(O(sqrt(a)log(a)) + a^2nm-time algorithm for the number a of edges not in the solution. Be robust! In this part of the thesis, we study the VECTOR CONNECTIVITY problem, where we are given a graph G, a vertex labeling ell from V(G) to {1, . . . , d }, and an integer k. We are to find a vertex subset S of V(G) of size at most k such that each vertex v in V (G)\S has ell(v) vertex-disjoint paths from v to S in G. Such a set S is useful when placing servers in a network to satisfy robustness-of-service demands. We prove that VECTOR CONNECTIVITY admits a randomized fixed-parameter algorithm with respect to k, that it does not allow a polynomial kernelization with respect to k + d but that, if d is treated as a constant, then it allows a vertex-linear kernelization with respect to k. In dieser Dissertation untersuchen wir die Berechnungskomplexität von fünf NP-schweren Graphproblemen. Es wird weithin angenommen, dass NP-schwere Probleme im Allgemeinen nicht effizient gelöst werden können, das heißt, dass sie keine Polynomialzeitalgorithmen erlauben. Diese Annahme basiert auf vielen bisher nicht erfolgreichen Versuchen das Gegenteil zu beweisen. Aus diesem Grund versuchen wir Eigenschaften der Eingabe herauszuarbeiten, die das betrachtete Problem handhabbar oder unhandhabbar machen. Solche Eigenschaften messen wir mittels Parametern, das heißt, Abbildungen, die jeder möglichen Eingabe eine natürliche Zahl zuordnen. Für einen gegebenen Parameter k versuchen wir dann Fixed-Parameter Algorithmen zu entwerfen, also Algorithmen, die auf Eingabe q eine obere Laufzeitschranke von f(k(q)) * |q|^c erlauben, wobei f eine, vorzugsweise schwach wachsende, Funktion ist, |q| die Länge der Eingabe, und c eine Konstante, vorzugsweise klein. In den Graphproblemen, die wir in dieser Dissertation studieren, repräsentiert unsere Eingabe eine Situation in der wir einen Lösungsgraph finden sollen. Zusätzlich sollen die Lösungsgraphen bestimmte problemspezifische Eigenschaften haben. Wir betrachten drei Varianten dieser Eigenschaften: Zunächst suchen wir einen Graphen, der sparse sein soll. Das heißt, dass er wenige Kanten enthalten soll. Dann suchen wir einen Graphen, der dense sein soll. Das heißt, dass er viele Kanten enthalten soll. Zuletzt suchen wir einen Graphen, der robust sein soll. Das heißt, dass er eine gute Lösung bleiben soll, selbst wenn er einige kleine Modifikationen durchmacht. Be sparse! In diesem Teil der Arbeit analysieren wir zwei ähnliche Probleme. In beiden ist die Eingabe ein Hypergraph H, bestehend aus einer Knotenmenge V und einer Familie E von Teilmengen von V, genannt Hyperkanten. Die Aufgabe ist einen Support für H zu finden, einen Graphen G, sodass für jede Hyperkante W in E der induzierte Teilgraph G[W] verbunden ist. Motiviert durch Anwendungen im Netzwerkdesign betrachten wir SUBSET INTERCONNECTION DESIGN, worin wir eine natürliche Zahl f als zusätzliche Eingabe bekommen, und der Support höchstens |V| - f + 1 Kanten enthalten soll. Wir zeigen, dass SUBSET INTERCONNECTION DESIGN einen Fixed-Parameter Algorithmus in Hinsicht auf die Zahl der Hyperkanten im Eingabegraph erlaubt, und einen Fixed-Parameter Algorithmus in Hinsicht auf f + d, wobei d die Größe einer größten Hyperkante ist. Motiviert durch eine Anwendung in der Hypergraphvisualisierung studieren wir r-OUTERPLANAR SUPPORT, worin der Support für H r-outerplanar sein soll, das heißt, er soll eine kantenkreuzungsfreie Einbettung in die Ebene erlauben mit höchstens r Schichten. Wir zeigen, dass r-OUTERPLANAR SUPPORT einen Fixed-Parameter Algorithmus in Hinsicht auf m + r zulässt, wobei m die Anzahl der Hyperkanten im Eingabehypergraphen H ist. Be dense! In diesem Teil der Arbeit studieren wir zwei Probleme, die durch Community Detection in sozialen Netzwerken motiviert sind. Dabei ist die Eingabe ein Graph G und eine natürliche Zahl k. Wir suchen einen Teilgraphen G' von G, der (genau) k Knoten enthält und dabei eine von zwei mathematisch präzisen Definitionen davon, dense zu sein, aufweist. In mu-CLIQUE, 0 < mu <= 1, soll der gesuchte Teilgraph G' mindestens mu mal k über 2 Kanten enthalten. Wir studieren die Berechnungskomplexität von mu-CLIQUE in Hinsicht auf drei Parameter des Eingabegraphen G: dem maximalen Knotengrad delta, dem h-Index h, und der Degeneracy d. Es gilt delta >= h >= d für jeden Graphen und h als auch d nehmen kleine Werte in Graphen an, die aus sozialen Netzwerken abgeleitet sind. Für delta und h erhalten wir Fixed-Parameter Algorithmen für mu-CLIQUE und wir zeigen, dass für d + k wahrscheinlich kein Fixed-Parameter Algorithmus existiert. Unsere positiven algorithmischen Resultate erhalten wir durch Entwickeln eines allgemeinen Frameworks zum Optimieren von Zielfunktionen über k-Knoten-Teilgraphen. In HIGHLY CONNECTED SUBGRAPH soll in dem gesuchten k-Knoten-Teilgraphen G' jeder Knoten Knotengrad mindestens floor(k/2) + 1 haben. Wir analysieren einen Teil der sogenannten Parameter Ecology für HIGHLY CONNECTED SUBGRAPH. Das heißt, wir navigieren im Raum der möglichen Parameter auf der Suche nach einem vernünftigen Trade-off zwischen kleinen Parameterwerten in der Praxis und effizienten oberen Laufzeitschranken. Die Highlights hier sind, dass es keine Algorithmen mit 2^o(n) * poly(n)-Laufzeit für HIGHLY CONNECTED SUBGRAPH gibt, es sei denn die Exponential Time Hypothesis stimmt nicht; die Entwicklung eines Algorithmus mit O(4^y * n^2 )-Laufzeit, wobei y die Anzahl der Kanten ist, die aus dem Lösungsgraphen G' herausgehen; und die Entwicklung eines Algorithmus mit 2^O(sqrt(a) log(a)) + O(a^2nm)-Laufzeit, wobei a die Anzahl der Kanten ist, die nicht in G' enthalten sind.
Computer Vision Eccv 2002
DOWNLOAD
Author : Anders Heyden
language : en
Publisher: Springer
Release Date : 2003-08-02
Computer Vision Eccv 2002 written by Anders Heyden and has been published by Springer this book supported file pdf, txt, epub, kindle and other format this book has been release on 2003-08-02 with Computers categories.
Premiering in 1990 in Antibes, France, the European Conference on Computer Vision, ECCV, has been held biennially at venues all around Europe. These conferences have been very successful, making ECCV a major event to the computer vision community. ECCV 2002 was the seventh in the series. The privilege of organizing it was shared by three universities: The IT University of Copenhagen, the University of Copenhagen, and Lund University, with the conference venue in Copenhagen. These universities lie ̈ geographically close in the vivid Oresund region, which lies partly in Denmark and partly in Sweden, with the newly built bridge (opened summer 2000) crossing the sound that formerly divided the countries. We are very happy to report that this year’s conference attracted more papers than ever before, with around 600 submissions. Still, together with the conference board, we decided to keep the tradition of holding ECCV as a single track conference. Each paper was anonymously refereed by three different reviewers. For the nal selection, for the rst time for ECCV, a system with area chairs was used. These met with the program chairsinLundfortwodaysinFebruary2002toselectwhatbecame45oralpresentations and 181 posters.Also at this meeting the selection was made without knowledge of the authors’identity.
Machine Learning And Knowledge Discovery In Databases Research Track
DOWNLOAD
Author : Albert Bifet
language : en
Publisher: Springer Nature
Release Date : 2024-08-29
Machine Learning And Knowledge Discovery In Databases Research Track written by Albert Bifet and has been published by Springer Nature this book supported file pdf, txt, epub, kindle and other format this book has been release on 2024-08-29 with Computers categories.
This multi-volume set, LNAI 14941 to LNAI 14950, constitutes the refereed proceedings of the European Conference on Machine Learning and Knowledge Discovery in Databases, ECML PKDD 2024, held in Vilnius, Lithuania, in September 2024. The papers presented in these proceedings are from the following three conference tracks: - Research Track: The 202 full papers presented here, from this track, were carefully reviewed and selected from 826 submissions. These papers are present in the following volumes: Part I, II, III, IV, V, VI, VII, VIII. Demo Track: The 14 papers presented here, from this track, were selected from 30 submissions. These papers are present in the following volume: Part VIII. Applied Data Science Track: The 56 full papers presented here, from this track, were carefully reviewed and selected from 224 submissions. These papers are present in the following volumes: Part IX and Part X.
New Developments In Statistical Information Theory Based On Entropy And Divergence Measures
DOWNLOAD
Author : Leandro Pardo
language : en
Publisher: MDPI
Release Date : 2019-05-20
New Developments In Statistical Information Theory Based On Entropy And Divergence Measures written by Leandro Pardo and has been published by MDPI this book supported file pdf, txt, epub, kindle and other format this book has been release on 2019-05-20 with Social Science categories.
This book presents new and original research in Statistical Information Theory, based on minimum divergence estimators and test statistics, from a theoretical and applied point of view, for different statistical problems with special emphasis on efficiency and robustness. Divergence statistics, based on maximum likelihood estimators, as well as Wald’s statistics, likelihood ratio statistics and Rao’s score statistics, share several optimum asymptotic properties, but are highly non-robust in cases of model misspecification under the presence of outlying observations. It is well-known that a small deviation from the underlying assumptions on the model can have drastic effect on the performance of these classical tests. Specifically, this book presents a robust version of the classical Wald statistical test, for testing simple and composite null hypotheses for general parametric models, based on minimum divergence estimators.
Adversarial Robustness For Machine Learning
DOWNLOAD
Author : Pin-Yu Chen
language : en
Publisher: Academic Press
Release Date : 2022-08-20
Adversarial Robustness For Machine Learning written by Pin-Yu Chen and has been published by Academic Press this book supported file pdf, txt, epub, kindle and other format this book has been release on 2022-08-20 with Computers categories.
Adversarial Robustness for Machine Learning summarizes the recent progress on this topic and introduces popular algorithms on adversarial attack, defense and verification. Sections cover adversarial attack, verification and defense, mainly focusing on image classification applications which are the standard benchmark considered in the adversarial robustness community. Other sections discuss adversarial examples beyond image classification, other threat models beyond testing time attack, and applications on adversarial robustness. For researchers, this book provides a thorough literature review that summarizes latest progress in the area, which can be a good reference for conducting future research. In addition, the book can also be used as a textbook for graduate courses on adversarial robustness or trustworthy machine learning. While machine learning (ML) algorithms have achieved remarkable performance in many applications, recent studies have demonstrated their lack of robustness against adversarial disturbance. The lack of robustness brings security concerns in ML models for real applications such as self-driving cars, robotics controls and healthcare systems. - Summarizes the whole field of adversarial robustness for Machine learning models - Provides a clearly explained, self-contained reference - Introduces formulations, algorithms and intuitions - Includes applications based on adversarial robustness
Robust Perception From Optical Sensors For Reactive Behaviors In Autonomous Robotic Vehicles
DOWNLOAD
Author : Alexander Schaub
language : en
Publisher: Springer
Release Date : 2017-07-18
Robust Perception From Optical Sensors For Reactive Behaviors In Autonomous Robotic Vehicles written by Alexander Schaub and has been published by Springer this book supported file pdf, txt, epub, kindle and other format this book has been release on 2017-07-18 with Technology & Engineering categories.
Alexander Schaub examines how a reactive instinctive behavior, similar to instinctive reactions as incorporated by living beings, can be achieved for intelligent mobile robots to extend the classic reasoning approaches. He identifies possible applications for reactive approaches, as they enable a fast response time, increase robustness and have a high abstraction ability, even though reactive methods are not universally applicable. The chosen applications are obstacle avoidance and relative positioning – which can also be utilized for navigation – and a combination of both. The implementation of reactive instinctive behaviors for the identified tasks is then validated in simulation together with real world experiments.
Energy Efficiency And Robustness Of Advanced Machine Learning Architectures
DOWNLOAD
Author : Alberto Marchisio
language : en
Publisher: CRC Press
Release Date : 2024-11-14
Energy Efficiency And Robustness Of Advanced Machine Learning Architectures written by Alberto Marchisio and has been published by CRC Press this book supported file pdf, txt, epub, kindle and other format this book has been release on 2024-11-14 with Computers categories.
Machine Learning (ML) algorithms have shown a high level of accuracy, and applications are widely used in many systems and platforms. However, developing efficient ML-based systems requires addressing three problems: energy-efficiency, robustness, and techniques that typically focus on optimizing for a single objective/have a limited set of goals. This book tackles these challenges by exploiting the unique features of advanced ML models and investigates cross-layer concepts and techniques to engage both hardware and software-level methods to build robust and energy-efficient architectures for these advanced ML networks. More specifically, this book improves the energy efficiency of complex models like CapsNets, through a specialized flow of hardware-level designs and software-level optimizations exploiting the application-driven knowledge of these systems and the error tolerance through approximations and quantization. This book also improves the robustness of ML models, in particular for SNNs executed on neuromorphic hardware, due to their inherent cost-effective features. This book integrates multiple optimization objectives into specialized frameworks for jointly optimizing the robustness and energy efficiency of these systems. This is an important resource for students and researchers of computer and electrical engineering who are interested in developing energy efficient and robust ML.
Business Intelligence And Information Technology
DOWNLOAD
Author : Aboul Ella Hassanien
language : en
Publisher: Springer Nature
Release Date : 2021-12-15
Business Intelligence And Information Technology written by Aboul Ella Hassanien and has been published by Springer Nature this book supported file pdf, txt, epub, kindle and other format this book has been release on 2021-12-15 with Technology & Engineering categories.
This book constitutes the refereed proceedings of the 2021 International Conference on Business Intelligence and Information Technology (BIIT 2021) held in Harbin, China, during December 18–20, 2021. BIIT 2021 is organized by the School of Computer and Information Engineering, Harbin University of Commerce, and supported by Scientific Research Group in Egypt (SRGE), Egypt. The papers cover current research in electronic commerce technology and application, business intelligence and decision making, digital economy, accounting informatization, intelligent information processing, image processing and multimedia technology, signal detection and processing, communication engineering and technology, information security, automatic control technique, data mining, software development, and design, blockchain technology, big data technology, artificial intelligence technology.
Quality Reliability Security And Robustness In Heterogeneous Systems
DOWNLOAD
Author : Xingliang Yuan
language : en
Publisher: Springer Nature
Release Date : 2021-11-16
Quality Reliability Security And Robustness In Heterogeneous Systems written by Xingliang Yuan and has been published by Springer Nature this book supported file pdf, txt, epub, kindle and other format this book has been release on 2021-11-16 with Computers categories.
This book constitutes the refereed post-conference proceedings of the 17th EAI International Conference on Quality, Reliability, Security and Robustness in Heterogeneous Networks, QShine 2021, held in November 2020. Due to COVID-19 pandemic the conference was held virtually. The 20 revised full papers were carefully reviewed and selected from 43 submissions. The papers are organized thematically in tracks Machine Learning in Distributed Networks; 5G Networks and Security; IoT Security and Lightweight Cryptography; Network Security; and Privacy-preserving Emerging Networked Applications.
Pattern Recognition And Machine Intelligence
DOWNLOAD
Author : Bhabesh Deka
language : en
Publisher: Springer Nature
Release Date : 2019-11-25
Pattern Recognition And Machine Intelligence written by Bhabesh Deka and has been published by Springer Nature this book supported file pdf, txt, epub, kindle and other format this book has been release on 2019-11-25 with Computers categories.
The two-volume set of LNCS 11941 and 11942 constitutes the refereed proceedings of the 8th International Conference on Pattern Recognition and Machine Intelligence, PReMI 2019, held in Tezpur, India, in December 2019. The 131 revised full papers presented were carefully reviewed and selected from 341 submissions. They are organized in topical sections named: Pattern Recognition; Machine Learning; Deep Learning; Soft and Evolutionary Computing; Image Processing; Medical Image Processing; Bioinformatics and Biomedical Signal Processing; Information Retrieval; Remote Sensing; Signal and Video Processing; and Smart and Intelligent Sensors.