Kaiserslautern - Fachbereich Informatik
Refine
Year of publication
- 1999 (267)
- 1996 (50)
- 1994 (49)
- 1995 (48)
- 1998 (38)
- 1997 (35)
- 2021 (27)
- 2016 (25)
- 2019 (24)
- 2022 (24)
- 1993 (22)
- 2015 (22)
- 2023 (22)
- 2001 (21)
- 2020 (21)
- 2007 (19)
- 2013 (18)
- 2018 (18)
- 2002 (17)
- 2003 (16)
- 2014 (15)
- 2012 (14)
- 1992 (13)
- 2000 (13)
- 2004 (12)
- 2006 (11)
- 2009 (11)
- 2008 (9)
- 2005 (8)
- 2017 (8)
- 1991 (7)
- 2010 (7)
- 2024 (7)
- 2011 (5)
- 1979 (2)
- 1980 (1)
- 1983 (1)
- 1990 (1)
Document Type
- Preprint (346)
- Doctoral Thesis (235)
- Report (139)
- Article (134)
- Master's Thesis (45)
- Study Thesis (13)
- Conference Proceeding (8)
- Bachelor Thesis (4)
- Part of a Book (2)
- Habilitation (2)
Has Fulltext
- yes (928)
Keywords
- AG-RESY (64)
- PARO (31)
- Case-Based Reasoning (20)
- Visualisierung (20)
- SKALP (16)
- CoMo-Kit (15)
- Fallbasiertes Schliessen (12)
- RODEO (12)
- Robotik (12)
- HANDFLEX (11)
Faculty / Organisational entity
Termination of Rewriting
(1994)
More and more, term rewriting systems are applied in computer science aswell as in mathematics. They are based on directed equations which may be used as non-deterministic functional programs. Termination is a key property for computing with termrewriting systems.In this thesis, we deal with different classes of so-called simplification orderings which areable to prove the termination of term rewriting systems. Above all, we focus on the problemof applying these termination methods to examples occurring in practice. We introduce aformalism that allows clear representations of orderings. The power of classical simplifica-tion orderings - namely recursive path orderings, path and decomposition orderings, Knuth-Bendix orderings and polynomial orderings - is improved. Further, we restrict these orderingssuch that they are compatible with underlying AC-theories by extending well-known methodsas well as by developing new techniques. For automatically generating all these orderings,heuristic-based algorithms are given. A comparison of these orderings with respect to theirpowers and their time complexities concludes the theoretical part of this thesis. Finally, notonly a detailed statistical evaluation of examples but also a brief introduction into the designof a software tool representing the integration of the specified approaches is given.
Orderings on polynomial interpretations of operators represent a powerful technique for proving thetermination of rewriting systems. One of the main problems of polynomial orderings concerns thechoice of the right interpretation for a given rewriting system. It is very difficult to develop techniquesfor solving this problem. Here, we present three new heuristic approaches: (i) guidelines for dealingwith special classes of rewriting systems, (ii) an algorithm for choosing appropriate special polynomialsas well as (iii) an extension of the original polynomial ordering which supports the generation ofsuitable interpretations. All these heuristics will be applied to examples in order to illustrate theirpractical relevance.
Planar force or pressure is a fundamental physical aspect during any people-vs-people and people-vs-environment activities and interactions. It is as significant as the more established linear and angular acceleration (usually acquired by inertial measurement units). There have been several studies involving planar pressure in the discipline of activity recognition, as reviewed in the first chapter. These studies have shown that planar pressure is a promising sensing modality for activity recognition. However, they still take a niche part in the entire discipline, using ad hoc systems and data analysis methods. Mostly these studies were not followed by further elaborative works. The situation calls for a general framework that can help push planar pressure sensing into the mainstream.
This dissertation systematically investigates using planar pressure distribution sensing technology for ubiquitous and wearable activity recognition purposes. We propose a generic Textile Pressure Mapping (TPM) Framework, which encapsulates (1) design knowledge and guidelines, (2) a multi-layered tool including hardware, software and algorithms, and (3) an ensemble of empirical study examples. Through validation with various empirical studies, the unified TPM framework covers the full scope of application recognition, including the ambient, object, and wearable subspaces.
The hardware part constructs a general architecture and implementations in the large-scale and mobile directions separately. The software toolkit consists of four heterogeneous tiers: driver, data processing, machine learning, visualization/feedback. The algorithm chapter describes generic data processing techniques and a unified TPM feature set. The TPM framework offers a universal solution for other researchers and developers to evaluate TPM sensing modality in their application scenarios.
The significant findings from the empirical studies have shown that TPM is a versatile sensing modality. Specifically, in the ambient subspace, a sports mat or carpet with TPM sensors embedded underneath can distinguish different sports activities or different people's gait based on the dynamic change of body-print; a pressure sensitive tablecloth can detect various dining actions by the force propagated from the cutlery through the plates to the tabletop. In the object subspace, swirl office chairs with TPM sensors under the cover can be used to detect the seater's real-time posture; TPM can be used to detect emotion-related touch interactions for smart objects, toys or robots. In the wearable subspace, TPM sensors can be used to perform pressure-based mechanomyography to detect muscle and body movement; it can also be tailored to cover the surface of a soccer shoe to distinguish different kicking angles and intensities.
All the empirical evaluations have resulted in accuracies well-above the chance level of the corresponding number of classes, e.g., the `swirl chair' study has classification accuracy of 79.5% out of 10 posture classes and in the `soccer shoe' study the accuracy is 98.8% among 17 combinations of angle and intensity.
The Analytic Blossom
(2001)
Blossoming is a powerful tool for studying and computing with Bézier and B-spline curves and surfaces - that is, for the investigation and analysis of polynomials and piecewise polynomials in geometric modeling. In this paper, we define a notion of the blossom for Poisson curves. Poisson curves are to analytic functions what Bézier curves are to polynomials - a representation adapted to geometric design. As in the polynomial setting, the blossom provides a simple, powerful, elegant and computationally meaningful way to analyze Poisson curves. Here, we
define the analytic blossom and interpret all the known algorithms for Poisson curves - subdivision, trimming, evaluation of the function and its derivatives, and conversion between the Taylor and the Poisson basis - in terms of this analytic blossom.
In this report we present a case study of employing goal-oriented heuristics whenproving equational theorems with the (unfailing) Knut-Bendix completion proce-dure. The theorems are taken from the domain of lattice ordered groups. It will bedemonstrated that goal-oriented (heuristic) criteria for selecting the next critical paircan in many cases significantly reduce the search effort and hence increase per-formance of the proving system considerably. The heuristic, goalADoriented criteriaare on the one hand based on so-called "measures" measuring occurrences andnesting of function symbols, and on the other hand based on matching subterms.We also deal with the property of goal-oriented heuristics to be particularly helpfulin certain stages of a proof. This fact can be addressed by using them in a frame-work for distributed (equational) theorem proving, namely the "teamwork-method".
We describe a platform for the portable and secure execution of mobile agents writtenin various interpreted languages on top of a common run-time core. Agents may migrate at anypoint in their execution, fully preserving their state, and may exchange messages with otheragents. One system may contain many virtual places, each establishing a domain of logicallyrelated services under a common security policy governing all agents at this place. Agents areequipped with allowances limiting their resource accesses, both globally per agent lifetime andlocally per place. We discuss aspects of this architecture and report about ongoing work.
The calculation of form factors is an important problem in computing the global illumination in the radiosity setting. Closed form solutions often are only available for objects without obstruction and are very hard to calculate. Using Monte Carlo integration and ray tracing provides a fast and elegant tool for the estimation of the form factors. In this paper we show, that using deterministic low discrepancy sample points is superior to random sampling, resulting in an acceleration of more than half an order of magnitude.
Virtual Reality (VR) is to be seen as the superset of simulation and animation. Visualization is done by rendering. The fundamental model of VR accounts for all phenomenons to be modelled with help of a computer. Examples range from simple dragging actions with a mouse device to the complex simulation of physically based animation.
Optimal degree reductions, i.e. best approximations of \(n\)-th degree Bezier curves
by Bezier curves of degree \(n\) - 1, with respect to different norms are studied. It
is shown that for any \(L_p\)-norm the euclidean degree reduction where the norm is applied to the euclidean distance function of two curves is identical to componentwise degree reduction. The Bezier points of the degree reductions are found to lie on parallel lines through the Bezier points of any Taylor expansion of degree \(n\) - 1 of the original curve. This geometric situation is shown to hold also in the case of constrained degree reduction. The Bezier points of the degree reduction are explicitly given in the unconstrained case for \(p\) = 1 and \(p\) = 2 and in the constrained case for \(p\) = 2.
This paper presents a brief overview of the INRECA-II methodology for building and maintaining CBR applications. It is based on the experience factory and the software process modeling approach from software engineering. CBR development and maintenance experience is documented using software process models and stored in a three-layered experience packet.
Load balancing is one of the central problems that have to be solved in parallel computation. Here, the problem of distributed, dynamic load balancing for massive parallelism is addressed. A new local method, which realizes a physical analogy to equilibrating liquids in multi-dimensional tori or hypercubes, is presented. It is especially suited for communication mechanisms with low set-up to transfer ratio occurring in tightly-coupled or SIMD systems. By successive shifting single load elements to the direct neighbors, the load is automatically transferred to lightly loaded processors. Compared to former methods, the proposed Liquid model has two main advantages. First, the task of load sharing is combined with the task of load balancing, where the former has priority. This property is valuable in many applications and important for highly dynamic load distribution. Second, the Liquid model has high efficiency. Asymptotically, it needs O(D . K . Ldiff ) load transfers to reach the balanced state in a D-dimensional torus with K processors per dimension and a maximum initial load difference of Ldiff . The Liquid model clearly outperforms an earlier load balancing approach, the nearest-neighbor-averaging. Besides a survey of related research, analytical results within a formal framework are derived. These results are validated by worst-case simulations in one-and two-dimensional tori with up to two thousand processors.
We consider two major topics in this thesis: spatial domain partitioning which serves as a framework to simulate creep flows in representative volume elements.
First, we introduce a novel multi-dimensional space partitioning method. A new type of tree combines the advantages of the Octree and the KD-tree without having their disadvantages. We present a new data structure allowing local refinement, parallelization and proper restriction of transition ratios between nodes. Our technique has no dimensional restrictions at all. The tree's data structure is defined by a topological algebra based on the symbols \( A = \{ L, I, R \} \) that encode the partitioning steps. The set of successors is restricted such that each node has the partition of unity property to partition domains without overlap. With our method it is possible to construct a wide choice of spline spaces to compress or reconstruct scientific data such as pressure and velocity fields and multidimensional images. We present a generator function to build a tree that represents a voxel geometry. The space partitioning system is used as a framework to allow numerical computations. This work is triggered by the problem of representing, in a numerically appropriate way, huge three-dimensional voxel geometries that could have up to billions of voxels. These large datasets occure in situations where it is needed to deal with large representative volume elements (REV).
Second, we introduce a novel approach of variable arrangement for pressure and velocity to solve the Stokes equations. The basic idea of our method is to arrange variables in a way such that each cell is able to satisfy a given physical law independently from its neighbor cells. This is done by splitting velocity values to a left and right converging component. For each cell we can set up a small linear system that describes the momentum and mass conservation equations. This formulation allows to use the Gauß-Seidel algorithm to solve the global linear system. Our tree structure is used for spatial partitioning of the geometry and provides a proper initial guess. In addition, we introduce a method that uses the actual velocity field to refine the tree and improve the numerical accuracy where it is needed. We developed a novel approach rather than using existing approaches such as the SIMPLE algorithm, Lattice-Boltzmann methods or Exlicit jump methods since they are suited for regular grid structures. Other standard CFD approaches extract surfaces and creates tetrahedral meshes to solve on unstructured grids thus can not be applied to our datastructure. The discretization converges to the analytical solution with respect to grid refinement. We conclude a high strength in computational time and memory for high porosity geometries and a high strength in memory requirement for low porosity geometries.
We present an empirical study of mathematical proofs by diagonalization, the aim istheir mechanization based on proof planning techniques. We show that these proofs canbe constructed according to a strategy that (i) finds an indexing relation, (ii) constructsa diagonal element, and (iii) makes the implicit contradiction of the diagonal elementexplicit. Moreover we suggest how diagonal elements can be represented.
Approximation properties of the underlying estimator are used to improve the efficiency of the method of dependent tests. A multilevel approximation procedure is developed such that in each level the number of samples is balanced with the level-dependent variance, resulting in a considerable reduction of the overall computational cost. The new technique is applied to the Monte Carlo estimation of integrals depending on a parameter.
Comprehensive reuse and systematic evolution of reuse artifacts as proposed by the Quality Improvement Paradigm (QIP) do not only require tool support for mere storage and retrieval. Rather, an integrated management of (potentially reusable) experience data as well as project-related data is needed. This paper presents an approach exploiting object-relational database technology to implement the QIP-driven reuse repository of the SFB 501. Requirements, concepts, and implementational aspects are discussed and illustrated through a running example, namely the reuse and continuous improvement of SDL patterns for developing distributed systems. Based on this discussion, we argue that object-relational database management systems (ORDBMS) are best suited to implement such a comprehensive reuse repository. It is demonstrated how this technology can be used to support all phases of a reuse process and the accompanying improvement cycle. Although the discussions of this paper are strongly related to the requirements of the SFB 501 experience base, the basic realization concepts, and, thereby, the applicability of ORDBMS, can easily be extended to similar applications, i. e., reuse repositories in general.
Most automated theorem provers suffer from the problem that theycan produce proofs only in formalisms difficult to understand even forexperienced mathematicians. Efforts have been made to transformsuch machine generated proofs into natural deduction (ND) proofs.Although the single steps are now easy to understand, the entire proofis usually at a low level of abstraction, containing too many tedioussteps. Therefore, it is not adequate as input to natural language gen-eration systems.To overcome these problems, we propose a new intermediate rep-resentation, called ND style proofs at the assertion level . After illus-trating the notion intuitively, we show that the assertion level stepscan be justified by domain-specific inference rules, and that these rulescan be represented compactly in a tree structure. Finally, we describea procedure which substantially shortens ND proofs by abstractingthem to the assertion level, and report our experience with furthertransformation into natural language.
The intuitionistic calculus mj for sequents, in which no other logical symbols than those for implication and universal quantification occur, is introduced and analysed. It allows a simple backward application, called mj-reduction here, for searching for derivation trees. Terms needed in mj-reduction can be found with the unification algorithm. mj-Reduction with unification can be seen as a natural extension of SLD-resolution. mj-Derivability of the sequents considered here coincides with derivability in Johansson's minimal intuitionistic calculus LHM in [6]. Intuitionistic derivability of formulae with negation and classical derivability of formulae with all usual logical symbols can be expressed with mj-derivability and hence be verified by mj-reduction. mj-Derivations can be easily translated into LJ-derivations without
"Schnitt", or into NJ-derivations in a slightly sharpened form of Prawitz' normal form. In the first three sections, the systematic use of mj-reduction for proving in predicate logic is emphasized. Although the fourth section, the last and largest, is exclusively devoted to the mathematical analysis of the calculus mj, the first three sections may be of interest to a wider readership, including readers looking for applications of symbolic logic. Unfortunately, the mathematical analysis of the calculus mj, as the study of Gentzen's calculi, demands a large amount of technical work that obscures the natural unfolding of the argumentation. To alleviate this, definitions and theorems are completely embedded in the text to provide a fluent and balanced mathematical discourse: new concepts are indicated with bold-face, proofs of assertions are outlined, or omitted when it is assumed that the reader can provide them.
Fluid extraction is a typical chemical process where two types of fluids are mixed together. The high complexity of this process which involves droplet coalescence, breakup, mass transfer, and counter-current flow often makes design difficult. The industrial design of these processes is still based on expensive mini-plant and pilot plant experiments. Therefore, there is a strong need for research into the stimulation of fluid-fluid interaction processes using computational fluid dynamics (CFD).
Previous multi-phase fluid simulations have focused on the development of models that couple mass and momentum using the Navier-Stokes equation. Recent population balance models (PBM) have proved to be important methods for analyzing droplet breakage and collisions. A combination of CFD and PBM facilitates the simulation of flow property by solving coupling equations, and the calculation of the droplet size and numbers. In our study, we successfully coupled an Euler-Euler CFD model with the breakup and coalescence models proposed by Luo and Svendsen (59).
The simulation output of extraction columns provides a mathematical understand- ing of how fluids are mixed inside a mixing device. This mixing process shows that the dispersed phase of a flow generates large blobs and bubbles. Current mathemati- cal simulation results often fail to provide an intuitive representation of how well two different types of fluid interact, so intuitive and physically plausible visualization tech- niques are in high demand to help chemical engineers to explore and analyze bubble column simulation data. In chapter 3, we present the visualization tools we developed for extraction column data.
Fluid interfaces and free surfaces are topics of growing interest in the field of multi- phase computational fluid dynamics. However, the analysis of the flow field relative to the material interface shape and topology is a challenging task. In chapter 5, we present a technique that facilitates the visualization and analysis of complex material interface behaviors over time. To achieve this, we track the surface parameterization of time-varying material interfaces and identify locations where there are interactions between the material interfaces and fluid particles. Splatting and surface visualization techniques produce an intuitive representation of the derived interface stability. Our results demonstrate that the interaction of a flow field with a material interface can be understood using appropriate extraction and visualization techniques, and that our techniques can help the analysis of mixing and material interface consistency.
In addition to texture-based methods for surface analysis, the interface of two- phase fluid can be considered as an implicit function of the density or volume fraction values. High-level visualization techniques such as topology-based methods can re- veal the hidden structure underlying simple simulation data, which will enhance and advance our understanding of multi-fluid simulation data. Recent feature-based vi- sualization approaches have explored the possibility of using Reeb graphs to analyze scalar field topologies(19, 107). In chapter 6, we present a novel interpolation scheme for interpolating point-based volume fraction data and we further explore the implicit fluid interface using a topology-based method.
This paper shows how a new approach to theorem provingby analogy is applicable to real maths problems. This approach worksat the level of proof-plans and employs reformulation that goes beyondsymbol mapping. The Heine-Borel theorem is a widely known result inmathematics. It is usually stated in R 1 and similar versions are also truein R 2 , in topology, and metric spaces. Its analogical transfer was proposedas a challenge example and could not be solved by previous approachesto theorem proving by analogy. We use a proof-plan of the Heine-Boreltheorem in R 1 as a guide in automatically producing a proof-plan of theHeine-Borel theorem in R 2 by analogy-driven proof-plan construction.
In this paper we are interested in an algebraic specification language that (1) allowsfor sufficient expessiveness, (2) admits a well-defined semantics, and (3) allows for formalproofs. To that end we study clausal specifications over built-in algebras. To keep thingssimple, we consider built-in algebras only that are given as the initial model of a Hornclause specification. On top of this Horn clause specification new operators are (partially)defined by positive/negative conditional equations. In the first part of the paper wedefine three types of semantics for such a hierarchical specification: model-theoretic,operational, and rewrite-based semantics. We show that all these semantics coincide,provided some restrictions are met. We associate a distinguished algebra A spec to ahierachical specification spec. This algebra is initial in the class of all models of spec.In the second part of the paper we study how to prove a theorem (a clause) valid in thedistinguished algebra A spec . We first present an abstract framework for inductive theoremprovers. Then we instantiate this framework for proving inductive validity. Finally wegive some examples to show how concrete proofs are carried out.This report was supported by the Deutsche Forschungsgemeinschaft, SFB 314 (D4-Projekt)
This technical report is the Emerging Trends proceedings of the 20th International Conference on Theorem Proving in Higher Order Logics (TPHOLs 2007), which was held during 10-13 September in Kaiserslautern, Germany. TPHOLs covers all aspects of theorem proving in higher order logics as well as related topics in theorem proving and verification.
Neuronale Netze sind ein derzeit (wieder) aktuelles Thema. Trotz der oft eher schlagwortartigen
Verwendung dieses Begriffs beinhaltet er eine Vielfalt von Ideen, unterschiedlichste methodische
Ansätze und konkrete Anwendungsmöglichkeiten. Die grundlegenden Vorstellungen sind dabei nicht neu, sondern haben eine mitunter recht lange Tradition in angrenzenden Disziplinen wie Biologie, Kybernetik , Mathematik und Physik . Vielversprechende Forschungsergebnisse der letzten Zeit haben dieses Thema wieder in den Mittelpunkt des Interesses gerückt und eine Vielzahl neuer Querbezüge zur Informatik und Neurobiologie sowie zu anderen, auf den ersten Blick weit entfernten Gebieten offenbart. Gegenstand des Forschungsgebiets Neuronale Netze ist dabei die Untersuchung und Konstruktion informationsverarbeitender Systeme, die sich aus vielen mitunter nur sehr primitiven, uniformen Einheiten zusammensetzen und deren wesentliches Verarbeitungsprinzip die Kommunikation zwischen diesen Einheiten ist, d.h. die Übertragung von Nachrichten oder Signalen. Ein weiteres
Charakteristikum dieser Systeme ist die hochgradig parallele Verarbeitung von Information innerhalb
des Systems. Neben der Modellierung kognitiver Prozesse und dem Interesse, wie das menschliche Gehirn komplexe kognitive Leistungen vollbringt, ist über das rein wissenschaftliche Interesse hinaus in zunehmendem Maße auch der konkrete Einsatz neuronaler Netze in verschiedenen technischen Anwendungsgebieten zu sehen. Der vorliegende Report beinhaltet die schriftlichen Ausarbeitungen der Teilnehmerinnen des Seminars Theorie und Praxis neuronaler Netze , das von der Arbeitsgruppe Richter im Sommersemester 1993 an der Universität Kaiserslautern veranstaltet wurde. Besonderer Wert wurde darauf gelegt, nicht nur die theoretischen Grundlagen neuronaler Netze zu behandeln, sondern auch deren Einsatz in der Praxis zu diskutieren. Die Themenauswahl spiegelt einen Teil des weiten Spektrums der Arbeiten auf diesem Gebiet wider. Ein Anspruch auf Vollständigkeit kann daher nicht erhoben werden. Insbesondere sei darauf verwiesen, daß für eine intensive, vertiefende Beschäftigung mit einem Thema auf die jeweiligen Originalarbeiten zurückgegriffen werden sollte. Ohne die Mitarbeit der Teilnehmerinnen und Teilnehmer des Seminars wäre dieser Report nicht möglich gewesen. Wir bedanken uns daher bei Frank Hauptmann, Peter Conrad, Christoph Keller, Martin Buch, Philip Ziegler, Frank Leidermann, Martin Kronenburg, Michael Dieterich, Ulrike Becker, Christoph Krome, Susanne Meyfarth , Markus Schmitz, Kenan Çarki, Oliver Schweikart, Michael Schick und Ralf Comes.
The wireless spectrum is already a scarce good, shared by multiple competing technologies such as Bluetooth, ZigBee and Wi-Fi, and the hunger for traffic is only increasing. Due to the heterogeneity of the existing wireless technologies and the real threat that interference poses to network performance, sophisticated techniques must be developed to ensure acceptable levels of quality of service.
In this thesis, we present a passive channel sensing scheme based on both energy and signal detection, that primarily considers the spectrum occupation of foreign traffic while allowing for additional complementary information such as the signal-to-noise ratio. The resulting channel quality metric is first corrected for the spectrum occupation of internal transmissions and later aggregated with help of a moving average followed by an exponential weighted moving average. This aggregation keeps the metric both sufficiently stable and adaptive to significant changes in channel usage. Moreover, the channel quality metric is made volatility-aware by penalizing qualities proportionally to their downward volatility. This yields a conservative metric and allows to differentiate channels with similar aggregated qualities but different volatility behavior.
Our second main contribution is in the form of a schedule-based channel sensing protocol, in which nodes possess two network interfaces, one for communication and one for channel sensing. Channel sensing schedules are derived from communication schedules, i.e. channel hopping sequences used for communication, with help of a stochastic local search-based heuristic that attempts to minimize channel sensing bias, the channel overlap between both schedules and to maximize overlap fairness. This minimizes the effect of internal transmissions in the resulting channel quality metric, allowing nodes to derive channel quality primarily based on foreign traffic in an unbiased manner.
Finally, we propose and implement a stabilization protocol for keeping nodes in an ad-hoc network tick-synchronized and schedule-consistent w.r.t. a communication schedule. This stabilization protocol makes use of special messages, namely tick frames for synchronization, channel quality reports for sharing local views of channel conditions and schedule reports for disseminating the global communication hopping sequence. The communication schedules are computed by a master node based on an aggregation of local channel quality views and the re-computation of these schedules is triggered by significant changes in channel conditions. The resulting protocol is robust against changes in topology and channel conditions.
During our daily lives, we are confronted with vast amounts of data, the processing of which can dramatically influence our lives, both positively and negatively. The enormous amount of data (images, texts, tables, and time series), its variety and possible applications are not always obvious. Due to advancements in the internet of things (IoT), there exist billions of sensors that produce time series which can be found everywhere, whether in medicine, the financial sector or the agricultural economy. This incredible amount of time series data has many hidden features which are useful for industry as well as for daily use, e.g. improving the cancer prediction can save real human lives. Recently, several deep learning methods have been proposed for analyzing this time series data. However, due to their black box nature, their applicability is limited in critical sectors like medicine, finance, and communication. In addition, it is now a compulsion as per artificial intelligence (AI) Act and per General Data Protection Regulation (GDPR) to protect any sensitive data and provide explanations in safety-critical domains. To enable use of DNNs in a broader domain scope, this thesis presents a framework for privacy-preserved and interpretable time series analysis. TimeFrame consists of four main components, namely, post-hoc interpretability, intrinsic interpretability, direct privacy, and indirect privacy. Interpretability is indispensable to avoid damaging people or the infrastructure. In the past years, the development mostly focused on image data, which prevented the full potential of DNNs in time series processing from being exploited. To overcome this limitation, TimeFrame introduces five (Time to Focus, TSViz, TimeREISE, TSInsight, Data Lens) novel post-hoc and two (PatchX, P2ExNet) novel intrinsic interpretability components. TimeFrame addresses multiple perspectives such as attribution, compression, visualization, influence, prototyping, and hierarchical splitting. Compared to existing methods, the components show better explanations, robustness, and scalability. Another crucial factor is the privacy when dealing with sensitive data and deep learning. In this context, TimeFrame introduces two (PPML, PPML x XAI) components for direct and one (From Private to Public) component for indirect privacy. These components benchmark privacy approaches, their effect on interpretability, and the synthetic generation of data to overcome privacy concerns. TimeFrame offers a large set of interpretability and privacy components that can be combined and consider numerous different aspects. Furthermore, the novel approaches have shown to consistently outperform twenty existing state-of-the-art methods across up to 20 different datasets. To guarantee the fairness, various metrics were used including performance change, Sensitivity, Infidelity, Continuity, runtime, model dependency, compression rate, and others. This broad set of metrics makes it possible to provide guidelines for a more appropriate use of existing state-of-the-art approaches as well as the novel components included in TimeFrame.
Today, information systems are often distributed to achieve high availability and low latency.
These systems can be realized by building on a highly available database to manage the distribution of data.
However, it is well known that high availability and low latency are not compatible with strong consistency guarantees.
For application developers, the lack of strong consistency on the database layer can make it difficult to reason about their programs and ensure that applications work as intended.
We address this problem from the perspective of formal verification.
We present a specification technique, which allows specifying functional properties of the application.
In addition to data invariants, we support history properties.
These let us express relations between events, including invocations of the application API and operations on the database.
To address the verification problem, we have developed a proof technique that handles concurrency using invariants and thereby reduces the problem to sequential verification.
The underlying system semantics, technique and its soundness proof are all formalized in the interactive theorem prover Isabelle/HOL.
Additionally, we have developed a tool named Repliss which uses the proof technique to enable partially automated verification and testing of applications.
For verification, Repliss generates verification conditions via symbolic execution and then uses an SMT solver to discharge them.
Today’s pervasive availability of computing devices enabled with wireless communication and location- or inertial sensing capabilities is unprecedented. The number of smartphones sold worldwide are still growing and increasing numbers of sensor enabled accessories are available which a user can wear in the shoe or at the wrist for fitness tracking, or just temporarily puts on to measure vital signs. Despite this availability of computing and sensing hardware the merit of application seems rather limited regarding the full potential of information inherent to such senor deployments. Most applications build upon a vertical design which encloses a narrowly defined sensor setup and algorithms specifically tailored to suit the application’s purpose. Successful technologies, however, such as the OSI model, which serves as base for internet communication, have used a horizontal design that allows high level communication protocols to be run independently from the actual lower-level protocols and physical medium access. This thesis contributes to a more horizontal design of human activity recognition systems at two stages. First, it introduces an integrated toolchain to facilitate the entire process of building activity recognition systems and to foster sharing and reusing of individual components. At a second stage, a novel method for automatic integration of new sensors to increase a system’s performance is presented and discussed in detail.
The integrated toolchain is built around an efficient toolbox of parametrizable components for interfacing sensor hardware, synchronization and arrangement of data streams, filtering and extraction of features, classification of feature vectors, and interfacing output devices and applications. The toolbox emerged as open-source project through several research projects and is actively used by research groups. Furthermore, the toolchain supports recording, monitoring, annotation, and sharing of large multi-modal data sets for activity recognition through a set of integrated software tools and a web-enabled database.
The method for automatically integrating a new sensor into an existing system is, at its core, a variation of well-established principles of semi-supervised learning: (1) unsupervised clustering to discover structure in data, (2) assumption that cluster membership is correlated with class membership, and (3) obtaining at a small number of labeled data points for each cluster, from which the cluster labels are inferred. In most semi-supervised approaches, however, the labels are the ground truth provided by the user. By contrast, the approach presented in this thesis uses a classifier trained on an N-dimensional feature space (old classifier) to provide labels for a few points in an (N+1)-dimensional feature space which are used to generate a new, (N+1)-dimensional classifier. The different factors that make a distribution difficult to handle are discussed, a detailed description of heuristics designed to mitigate the influences of such factors is provided, and a detailed evaluation on a set of over 3000 sensor combinations from 3 multi-user experiments that have been used by a variety of previous studies of different activity recognition methods is presented.
Using an experience factory is one possible concept for supporting and improving reuse in software development. (i.e., reuse of products, processes, quality models, ...). In the context of the Sonderforschungsbereich 501: "Development of Large Systems with Generic methods" (SFB501), the Software Engineering Laboratory (SE Lab) runs such an experience factory as part of the infrastructure services it offers. The SE Lab also provides several tools to support the planning, developing, measuring, and analyzing activities of software development processes. Among these tools, the SE Lab runs and maintains an experience base, the SFB-EB. When an experience factory is utilized, support for experience base maintenance is an important issue. Furthermore, it might be interesting to evaluate experience base usage with regard to the number of accesses to certain experience elements stored in the database. The same holds for the usage of the tools provided by the SE LAB. This report presents a set of supporting tools that were designed to aid in these tasks. These supporting tools check the experience base's consistency and gather information on the usage of SFB-EB and the tools installed in the SE Lab. The results are processed periodically and displayed as HTML result reports (consistency checking) or bar charts (usage profiles).
To increase situational awareness of the crane operator, the aim of this thesis is to develop a vision-based deep learning object detection from crane load-view using an adaptive perception in the construction area. Conventional worker detection methods are based on simple shape or color features from the worker's appearances. Nonetheless, these methods can fail to recognize the workers who do not wear the protective gears. To find out an image representation of the object from the top view manually or handcrafted feature is crucial. We, therefore, employed deep learning methods to automatically learn those features.
To yield optimal results, deep learning methods require mass amount of data.
Due to the data deficit especially in the construction domain, we developed the photorealistic world to create the data in addition to our samples collected from the real construction area. The simulated platform does not benefit only from diverse data types, but also concurrent research development which speeds up the pipeline at a low cost.
Our research findings indicate that the combination of synthetic and real training samples improved the state-of-the-art detector. In line with previous studies to bridge the gap between synthetic and real data, the results of preprocessed synthetic images are substantially better than using the raw data by approximately 10%.
Finding the right deep learning model for load-view detection is challenging.
By investigating our training data, it becomes evident that the majority of bounding box sizes are very small with a complex background.
In addition, we gave the priority to speed over accuracy based on the construction safety criteria. Finally, RetinaNet is chosen out of the three primary object detection models.
Nevertheless, the data-driven detection algorithm can fail to handle scale invariance, especially for detectors whose input size is changed in an extremely wide range.
The adaptive zoom feature can enhance the quality of the worker detection.
To avoid further data gathering and extensive retraining, the proposed automatic zoom method of the load-view crane camera supports the deep learning algorithm, specifically in the high scale variant problem. The finite state machine is employed for control strategies to adapt the zoom level to cope not only with inconsistent detection but also abrupt camera movement during lifting operation. Consequently, the detector is able to detect a small size object by smooth continuous zoom control without additional training.
The adaptive zoom control not only enhances the performance of the top-view object detection but also reduces the interaction of the crane operator with camera system, reducing the risk of fatality during load lifting operation.
Topology-Based Characterization and Visual Analysis of Feature Evolution in Large-Scale Simulations
(2019)
This manuscript presents a topology-based analysis and visualization framework that enables the effective exploration of feature evolution in large-scale simulations. Such simulations pose additional challenges to the already complex task of feature tracking and visualization, since the vast number of features and the size of the simulation data make it infeasible to naively identify, track, analyze, render, store, and interact with data. The presented methodology addresses these issues via three core contributions. First, the manuscript defines a novel topological abstraction, called the Nested Tracking Graph (NTG), that records the temporal evolution of features that exhibit a nesting hierarchy, such as superlevel set components for multiple levels, or filtered features across multiple thresholds. In contrast to common tracking graphs that are only capable of describing feature evolution at one hierarchy level, NTGs effectively summarize their evolution across all hierarchy levels in one compact visualization. The second core contribution is a view-approximation oriented image database generation approach (VOIDGA) that stores, at simulation runtime, a reduced set of feature images. Instead of storing the features themselves---which is often infeasable due to bandwidth constraints---the images of these databases can be used to approximate the depicted features from any view angle within an acceptable visual error, which requires far less disk space and only introduces a neglectable overhead. The final core contribution combines these approaches into a methodology that stores in situ the least amount of information necessary to support flexible post hoc analysis utilizing NTGs and view approximation techniques.
This paper focuses on the issues involved when multiple mobile agents interact in multiagent systems. The application is an intelligent agent market place, where buyer and seller agents cooperate and compete to process sales transactions for their owners. The market place manager acts as afacilitator by giving necessary information to agents and managing communication between agents, and also as a mediator by proposing solutions to agents or stopping them to get into infinite loops bargaining back and forth.The buyer and seller agents range from using hardcoded logic to rule-based inferencing in their negotiation strategies. However these agents must support some communication skills using KQML or FIPA-ACL.So in contrast with other approaches to multiagent negotiation, we introduce an explicit mediator (market place manager) into the negotiation, and we propose a negotiation strategy based on dependence theory [1] implemented by our best buyers and best sellers.
Background: The use of health apps to support the treatment of chronic pain is gaining importance. Most available pain management apps are still lacking in content quality and quantity as their developers neither involve health experts to ensure target group suitability nor use gamification to engage and motivate the user. To close this gap, we aimed to develop a gamified pain management app, Pain-Mentor.
Objective: To determine whether medical professionals would approve of Pain-Mentor’s concept and content, this study aimed to evaluate the quality of the app’s first prototype with experts from the field of chronic pain management and to discover necessary improvements.
Methods: A total of 11 health professionals with a background in chronic pain treatment and 2 mobile health experts participated in this study. Each expert first received a detailed presentation of the app. Afterward, they tested Pain-Mentor and then rated its quality using the mobile application rating scale (MARS) in a semistructured interview.
Results: The experts found the app to be of excellent general (mean 4.54, SD 0.55) and subjective quality (mean 4.57, SD 0.43). The app-specific section was rated as good (mean 4.38, SD 0.75). Overall, the experts approved of the app’s content, namely, pain and stress management techniques, behavior change techniques, and gamification. They believed that the use of gamification in Pain-Mentor positively influences the patients’ motivation and engagement and thus has the potential to promote the learning of pain management techniques. Moreover, applying the MARS in a semistructured interview provided in-depth insight into the ratings and concrete suggestions for improvement.
Conclusions: The experts rated Pain-Mentor to be of excellent quality. It can be concluded that experts perceived the use of gamification in this pain management app in a positive manner. This showed that combining pain management with gamification did not negatively affect the app’s integrity. This study was therefore a promising first step in the development of Pain-Mentor.
The Basic Reference Model of ODP introduces a number of basic concepts in order to provide a common basis for the development of a coherent set of standards. To achieve this objective, a clear understanding of the basic concepts is one prerequisite. This paper makes an effort at clarifying some of the basic concepts independently of standardized or non-standardized formal description techniques. Among the basic concepts considered here are: agent, action, interaction, interaction point, architecture, behaviour, system, composition, refinement, and abstraction. In a case study, it is then shown how these basic concepts can be represented in a formal specification written in temporal logic.
Although several systematic analyses of existing approaches to adaptation have been published recently, a general formal adaptation framework is still missing. This paper presents a step into the direction of developing such a formal model of transformational adaptation. The model is based on the notion of the quality of a solution to a problem, while quality is meant in a more general sense and can also denote some kind of appropriateness, utility, or degree of correctness. Adaptation knowledge is then defined in terms of functions transforming one case into a successor case. The notion of quality provides us with a semantics for adaptation knowledge and allows us to define terms like soundness, correctness and completeness. In this view, adaptation (and even the whole CBR process) appears to be a special instance of an optimization problem.
Towards A Non-tracking Web
(2016)
Today, many publishers (e.g., websites, mobile application developers) commonly use third-party analytics services and social widgets. Unfortunately, this scheme allows these third parties to track individual users across the web, creating privacy concerns and leading to reactions to prevent tracking via blocking, legislation and standards. While improving user privacy, these efforts do not consider the functionality third-party tracking enables publishers to use: to obtain aggregate statistics about their users and increase their exposure to other users via online social networks. Simply preventing third-party tracking without replacing the functionality it provides cannot be a viable solution; leaving publishers without essential services will hurt the sustainability of the entire ecosystem.
In this thesis, we present alternative approaches to bridge this gap between privacy for users and functionality for publishers and other entities. We first propose a general and interaction-based third-party cookie policy that prevents third-party tracking via cookies, yet enables social networking features for users when wanted, and does not interfere with non-tracking services for analytics and advertisements. We then present a system that enables publishers to obtain rich web analytics information (e.g., user demographics, other sites visited) without tracking the users across the web. While this system requires no new organizational players and is practical to deploy, it necessitates the publishers to pre-define answer values for the queries, which may not be feasible for many analytics scenarios (e.g., search phrases used, free-text photo labels). Our second system complements the first system by enabling publishers to discover previously unknown string values to be used as potential answers in a privacy-preserving fashion and with low computation overhead for clients as well as servers. These systems suggest that it is possible to provide non-tracking services with (at least) the same functionality as today’s tracking services.
In order to improve the quality of software systems and to set up a more effective process for their development, many attempts have been made in the field of software engineering. Reuse of existing knowledge is seen as a promising way to solve the outstanding problems in this field. In previous work we have integrated the design pattern concept with the formal design language SDL, resulting in a certain kind of pattern formalization. For the domain of communication systems we have also developed a pool of SDL patterns with an accompanying process model for pattern application. In this paper we present an extension that combines the SDL pattern approach with the experience base concept. This extension supports a systematic method for empirical evaluation and continuous improvement of the SDL pattern approach. Thereby the experience base serves as a repository necessary for effective reuse of the captured knowledge. A comprehensive usage scenario is described which shows the advantages of the combined approach. To demonstrate its feasibility, first results of a research case study are given.
Patients after total hip arthroplasty (THA) suffer from lingering musculoskeletal restrictions. Three-dimensional (3D) gait analysis in combination with machine-learning approaches is used to detect these impairments. In this work, features from the 3D gait kinematics, spatio temporal parameters (Set 1) and joint angles (Set 2), of an inertial sensor (IMU) system are proposed as an input for a support vector machine (SVM) model, to differentiate impaired and non-impaired gait. The features were divided into two subsets. The IMU-based features were validated against an optical motion capture (OMC) system by means of 20 patients after THA and a healthy control group of 24 subjects. Then the SVM model was trained on both subsets. The validation of the IMU system-based kinematic features revealed root mean squared errors in the joint kinematics from 0.24° to 1.25°. The validity of the spatio-temporal gait parameters (STP) revealed a similarly high accuracy. The SVM models based on IMU data showed an accuracy of 87.2% (Set 1) and 97.0% (Set 2). The current work presents valid IMU-based features, employed in an SVM model for the classification of the gait of patients after THA and a healthy control. The study reveals that the features of Set 2 are more significant concerning the classification problem. The present IMU system proves its potential to provide accurate features for the incorporation in a mobile gait-feedback system for patients after THA.
Recommender systems recommend items (e.g., movies, products, books) to users. In this thesis, we proposed two comprehensive and cluster-induced recommendation-based methods: Orthogonal Inductive Matrix Completion (OMIC) and Burst-induced Multi-armed Bandit (BMAB). Given the presence of side information, the first method is categorized as context-aware. OMIC is the first matrix completion method to approach the problem of incorporating biases, side information terms and a pure low-rank term into a single flexible framework with a well-principled optimization procedure. The second method, BMAB, is context-free. That is, it does not require any side data about users or items. Unlike previous context-free multi-armed bandit approaches, our method considers the temporal dynamics of human communication on the web and treats the problem in a continuous time setting. We built our models' assumptions under solid theoretical foundations. For OMIC, we provided theoretical guarantees in the form of generalization bounds by considering the distribution-free case: no assumptions about the sampling distribution are made. Additionally, we conducted a theoretical analysis of community side information when the sampling distribution is known and an adjusted nuclear norm regularization is applied. We showed that our method requires just a few entries to accurately recover the ratings matrix if the structure of the ground truth closely matches the cluster side information. For BMAB, we provided regret guarantees under mild conditions that demonstrate how the system's stability affects the expected reward. Furthermore, we conducted extensive experiments to validate our proposed methodologies. In a controlled environment, we implemented synthetic data generation techniques capable of replicating the domains for which OMIC and BMAB were designed. As a result, we were able to analyze our algorithms' performance across a broad spectrum of ground truth regimes. Finally, we replicated a real-world scenario by utilizing well-established recommender datasets. After comparing our approaches to several baselines, we observe that they achieved state-of-the-art results in terms of accuracy. Apart from being highly accurate, these methods improve interpretability by describing and quantifying features of the datasets they characterize.
To support scientific work with large and complex data the field of scientific visualization emerged in computer science and produces images through computational analysis of the data. Frameworks for combination of different analysis and visualization modules allow the user to create flexible pipelines for this purpose and set the standard for interactive scientific visualization used by domain scientists.
Existing frameworks employ a thread-parallel message-passing approach to parallel and distributed scalability, leaving the field of scientific visualization in high performance computing to specialized ad-hoc implementations. The task-parallel programming paradigm proves promising to improve scalability and portability in high performance computing implementations and thus, this thesis aims towards the creation of a framework for distributed, task-based visualization modules and pipelines.
The major contribution of the thesis is the establishment of modules for Merge Tree construction and (based on the former) topological simplification. Such modules already form a necessary first step for most visualization pipelines and can be expected to increase in importance for larger and more complex data produced and/or analysed by high performance computing.
To create a task-parallel, distributed Merge Tree construction module the construction process has to be completely revised. We derive a novel property of Merge Tree saddles and introduce a novel task-parallel, distributed Merge Tree construction method that has both good performance and scalability. This forms the basis for a module for topological simplification which we extend by introducing novel alternative simplification parameters that aim to reduce the importance of prior domain knowledge to increase flexibility in typical high performance computing scenarios.
Both modules lay the groundwork for continuative analysis and visualization steps and form a fundamental step towards an extensive task-parallel visualization pipeline framework for high performance computing.
Visualization is vital to the scientific discovery process.
An interactive high-fidelity rendering provides accelerated insight into complex structures, models and relationships.
However, the efficient mapping of visualization tasks to high performance architectures is often difficult, being subject to a challenging mixture of hardware and software architectural complexities in combination with domain-specific hurdles.
These difficulties are often exacerbated on heterogeneous architectures.
In this thesis, a variety of ray casting-based techniques are developed and investigated with respect to a more efficient usage of heterogeneous HPC systems for distributed visualization, addressing challenges in mesh-free rendering, in-situ compression, task-based workload formulation, and remote visualization at large scale.
A novel direct raytracing scheme for on-the-fly free surface reconstruction of particle-based simulations using an extended anisoptropic kernel model is investigated on different state-of-the-art cluster setups.
The versatile system renders up to 170 million particles on 32 distributed compute nodes at close to interactive frame rates at 4K resolution with ambient occlusion.
To address the widening gap between high computational throughput and prohibitively slow I/O subsystems, in situ topological contour tree analysis is combined with a compact image-based data representation to provide an effective and easy-to-control trade-off between storage overhead and visualization fidelity.
Experiments show significant reductions in storage requirements, while preserving flexibility for exploration and analysis.
Driven by an increasingly heterogeneous system landscape, a flexible distributed direct volume rendering and hybrid compositing framework is presented.
Based on a task-based dynamic runtime environment, it enables adaptable performance-oriented deployment on various platform configurations.
Comprehensive benchmarks with respect to task granularity and scaling are conducted to verify the characteristics and potential of the novel task-based system design.
A core challenge of HPC visualization is the physical separation of visualization resources and end-users.
Using more tiles than previously thought reasonable, a distributed, low-latency multi-tile streaming system is demonstrated, being able to sustain a stable 80 Hz when streaming up to 256 synchronized 3840x2160 tiles and achieve 365 Hz at 3840x2160 for sort-first compositing over the internet, thereby enabling lightweight visualization clients and leaving all the heavy lifting to the remote supercomputer.
We present first steps towards fully automated deduction that merely requiresthe user to submit proof problems and pick up results. Essentially, this necessi-tates the automation of the crucial step in the use of a deduction system, namelychoosing and configuring an appropriate search-guiding heuristic. Furthermore,we motivate why learning capabilities are pivotal for satisfactory performance.The infrastructure for automating both the selection of a heuristic and integra-tion of learning are provided in form of an environment embedding the "core"deduction system.We have conducted a case study in connection with a deduction system basedon condensed detachment. Our experiments with a fully automated deductionsystem 'AutoCoDe' have produced remarkable results. We substantiate Au-toCoDe's encouraging achievements with a comparison with the renowned the-orem prover Otter. AutoCoDe outperforms Otter even when assuming veryfavorable conditions for Otter.
The rising demand for machine learning (ML) models has become a growing concern for stakeholders who depend on automatic decisions. In today's world, black-box solutions (in particular deep neural networks) are being continuously implemented for more and more high-stake scenarios like medical diagnosis or autonomous vehicles. Unfortunately, when these opaque models make predictions that do not align with our expectations, finding a valid justification is simply not possible.
Explainable Artificial Intelligence (XAI) has emerged in response to our need for finding reasons that justify what a machine sees, but we don't. However, contributions in this field are mostly centered around local structures such as individual neurons or single input samples. Global characteristics that govern the behavior of a model are still poorly understood or have not been explored yet. An aggravating factor is the lack of a standard terminology to contextualize and compare contributions in this field. Such lack of consensus is depriving the ML community from ultimately moving away from black-boxes, and start creating systematic methods to design models that are interpretable by design.
So, what are the global patterns that govern the behavior of modern neural networks, and what can we do to make these models more interpretable from the start?
This thesis delves into both issues, unveiling patterns about existing models, and establishing strategies that lead to more interpretable architectures. These include biases coming from imbalanced datasets, quantification of model capacity, and robustness against adversarial attacks. When looking for new models that are interpretable by design, this work proposes a strategy to add more structure to neural networks, based on auxiliary tasks that are semantically related to the main objective. This strategy is the result of applying a novel theoretical framework proposed as part of this work. The XAI framework is meant to contextualize and compare contributions in XAI by providing actionable definitions for terms like "explanation" and "interpretation."
Altogether, these contributions address dire demands for understanding more about the global behavior of modern deep neural networks. More importantly, they can be used as a blueprint for designing novel, and more interpretable architectures. By tackling issues from the present and the future of XAI, results from this work are a firm step towards more interpretable models for computer vision.
Towards PACE-CAD Systems
(2022)
Despite phenomenal advancements in the availability of medical image datasets and the development of modern classification algorithms, Computer-Aided Diagnosis (CAD) has had limited practical exposure in the real-world clinical workflow. This is primarily because of the inherently demanding and sensitive nature of medical diagnosis that can have far-reaching and serious repercussions in case of misdiagnosis. In this work, a paradigm called PACE (Pragmatic, Accurate, Confident, & Explainable) is presented as a set of some of must-have features for any CAD. Diagnosis of glaucoma using Retinal Fundus Images (RFIs) is taken as the primary use case for development of various methods that may enrich an ordinary CAD system with PACE. However, depending on specific requirements for different methods, other application areas in ophthalmology and dermatology have also been explored.
Pragmatic CAD systems refer to a solution that can perform reliably in day-to-day clinical setup. In this research two, of possibly many, aspects of a pragmatic CAD are addressed. Firstly, observing that the existing medical image datasets are small and not representative of images taken in the real-world, a large RFI dataset for glaucoma detection is curated and published. Secondly, realising that a salient attribute of a reliable and pragmatic CAD is its ability to perform in a range of clinically relevant scenarios, classification of 622 unique cutaneous diseases in one of the largest publicly available datasets of skin lesions is successfully performed.
Accuracy is one of the most essential metrics of any CAD system's performance. Domain knowledge relevant to three types of diseases, namely glaucoma, Diabetic Retinopathy (DR), and skin lesions, is industriously utilised in an attempt to improve the accuracy. For glaucoma, a two-stage framework for automatic Optic Disc (OD) localisation and glaucoma detection is developed, which marked new state-of-the-art for glaucoma detection and OD localisation. To identify DR, a model is proposed that combines coarse-grained classifiers with fine-grained classifiers and grades the disease in four stages with respect to severity. Lastly, different methods of modelling and incorporating metadata are also examined and their effect on a model's classification performance is studied.
Confidence in diagnosing a disease is equally important as the diagnosis itself. One of the biggest reasons hampering the successful deployment of CAD in the real-world is that medical diagnosis cannot be readily decided based on an algorithm's output. Therefore, a hybrid CNN architecture is proposed with the convolutional feature extractor trained using point estimates and a dense classifier trained using Bayesian estimates. Evaluation on 13 publicly available datasets shows the superiority of this method in terms of classification accuracy and also provides an estimate of uncertainty for every prediction.
Explainability of AI-driven algorithms has become a legal requirement after Europe’s General Data Protection Regulations came into effect. This research presents a framework for easy-to-understand textual explanations of skin lesion diagnosis. The framework is called ExAID (Explainable AI for Dermatology) and relies upon two fundamental modules. The first module uses any deep skin lesion classifier and performs detailed analysis on its latent space to map human-understandable disease-related concepts to the latent representation learnt by the deep model. The second module proposes Concept Localisation Maps, which extend Concept Activation Vectors by locating significant regions corresponding to a learned concept in the latent space of a trained image classifier.
This thesis probes many viable solutions to equip a CAD system with PACE. However, it is noted that some of these methods require specific attributes in datasets and, therefore, not all methods may be applied on a single dataset. Regardless, this work anticipates that consolidating PACE into a CAD system can not only increase the confidence of medical practitioners in such tools but also serve as a stepping stone for the further development of AI-driven technologies in healthcare.
Mechanical ventilation of patients with severe lung injury is an important clinical treatment to ensure proper lung oxygenation and to mitigate the extent of collapsed lung regions. While current imaging technologies such as Computed Tomography (CT) and chest X-ray allow for a thorough inspection of the thorax, they are limited to static pictures and exhibit several disadvantages, including exposure to ionizing radiation and high cost. Electrical Impedance Tomography (EIT) is a novel method to determine functional processes inside the thorax such as lung ventilation and cardiac activity. EIT reconstructs the internal electrical conductivity distribution within the thorax from voltage measurements on the body surface. Conductivity changes correlate with important clinical parameters such as lung volume and perfusion. Current EIT systems and algorithms use simplified or generalized thorax models to solve the reconstruction problem, which reduce image quality and anatomical significance. In this thesis, the development of a clinically relevant workflow to compute sophisticated three-dimensional thorax models from patient-specific CT data is described. The method allows medical experts to generate a multi-material segmentation in an interactive and fast way, while a volumetric mesh is computed automatically from the segmentation. The significantly improved image quality and anatomical precision of EIT images reconstructed with these 3D models is reported, and the impact on clinical applicability is discussed. In addition, three projects concerning quantitative CT (qCT) measurements and multi-modal 3D visualization are presented, which demonstrate the importance and productivity of interdisciplinary research groups including computer scientists and medical experts. The results presented in this thesis contribute significantly to clinical research efforts to pave the way towards improved patient-specific treatments of lung injury using EIT and qCT.