FCT: DI - Teses de Doutoramento
URI permanente para esta coleção:
Navegar
Entradas recentes
- Information Design for Explainable Artificial Intelligence: Contributions to visualizing the Perceptron's decision-making process for non-expertsPublication . Heleno, Marco Jorge; Correia, Nuno; Fonseca, JoséThis study investigates how Information Design can contribute to Explainable Artificial Intelli- gence for non-expert audiences. Through an interdisciplinary and practice-based exploration, we focused on the Perceptron system, investigating which factors should be considered when design- ing visualizations for non-experts and which internal procedures challenge visual explainability. We developed and iterated through six visualizations to inform a seventh visualization. This last visualization was validated through semi-structured interviews and Think-Aloud sessions with non-experts. The results demonstrated how Information Design can contribute to the explainabil- ity of Artificial Intelligence, as participants developed a perceived understanding of the Percep- tron’s decision-making process without requiring specialized knowledge. Artificial Intelligence research has been iterating on a sub-field of Machine Learning known as Artificial Neural Networks. These developments add capabilities to these systems, but also increase the number of calculations involved in their inner workings. This creates an emerging statistical complexity that turns these systems into black boxes. Despite this incomprehensibility, the capabilities of these systems resulted in their wide- spread deployment, with their ubiquity causing a significant impact on a growing number of areas of society. This progress has raised concerns among researchers and policymakers who demand explanations of the decision-making process of these systems to ensure trustworthy interactions for users. The referred opacity, combined with the need for transparency, has led to the emergence of Explainable Artificial Intelligence. However, the literature shows that there is a lack of vis- ual explainers that communicate effectively with non-experts. These systems are based on net- works of artificial neurons, which are a fundamental building block for Artificial Neural Net- works. For this reason, they show to be an interesting entry point for understanding the basic mechanisms underlying modern machine learning systems, with the Perceptron being the model that most closely resembles this processing unit. Since Information Design has shown to be effective in visualizing and explaining dense and complex data sets, this research hypothesized that this approach could also help make the Perceptron's decision-making process intelligible for non-experts. This dissertation presents a validated approach to visualizing the Perceptron's decision- making process through a mechanical analogy. The visualization employs interactive elements, a retractable diagonal line encoding weighted sums, color-coded feedback revealing causal re- lationships between decisions and learning procedures, and an interactive iteration histogram. This approach demonstrates how visual abstractions can effectively reveal artificial decision- making processes for non-expert audiences.
- Scalable Consistency for Data ReplicationPublication . Fouto, Pedro Filipe Veiga; Leitão, João; Preguiça, NunoDistributed data storage solutions are key components of large-scale Internet services. The consistency guarantees provided by the protocols used to replicate data in such solu- tions can vary greatly, with some protocols providing strong guarantees while sacrificing availability, and others providing weaker guarantees while allowing higher availability. The choice of the replication protocol to use is often tied to the physical distribution of data replicas. As a general rule for Internet services, data replicated within a single datacen- ter is often replicated using protocols providing stronger consistency guarantees, while data replicated across multiple datacenters (i.e., geo-replicated) is often replicated using protocols providing weaker consistency guarantees. However, designing performant and fault-tolerant data replication solutions with data consistency that can scale in number of replicas is a challenging task. This is true not only for strong consistency models, but also for weaker consistency models, such as causal (and causal+) consistency. In this thesis, we propose to address the scalability and fault-tolerance limitations of providing consistency guarantees in data replication solutions, by addressing the entire spectrum of replication deployment scenarios, from single datacenter deployments, to geo-replicated, and edge computing deployments. To this end, we propose three main contributions, one for each of these deployment scenarios. First, we propose a new state-machine replication protocol, based on a new variant of Paxos, which improves on existing solutions by maximizing the throughput while avoiding performance degrada- tion when increasing the number of replicas. Second, we leverage the properties of the proposed protocol to replicate key components in geo-replicated causal consistency solu- tions, overcoming their fault-tolerance limitations while maximizing their performance. Third, by addressing the challenges of the edge computing environment, we propose a causal consistency data management solution that can efficiently scale to hundreds of edge locations.
- Intelligent Techniques for Interactive Video ContentPublication . Rodrigues, Rui Filipe dos Santos; Correia, Nuno; Madeira, RuiVideo can be used in many contexts, such as capturing important events, entertainment, or producing educational content. Moreover, video annotation and interactive video enable the viewer to actively engage with the video by providing various annotations or interactive elements, such as clickable buttons, hotspots, text, sketches, or other interactive features that enhance the viewing experience. This is especially relevant in contexts where users need to analyze specific types of video footage and highlight moments of interest. However, those highlights may not be easy to identify or interpret, often requiring significant time and expertise, and can lead to errors or overlooked information. Annotation systems can contribute to video analysis and understanding. These annotations enable the identification of significant elements in videos that require interpretation and creativity. These elements can serve as raw material for creating more detailed analyses, storytelling, or educational tools. Furthermore, these systems can facilitate collaboration, as multiple users can contribute annotations and insights, fostering a richer understanding of the video content. This work began with the development of a video annotation tool, which was subsequently refined and enhanced through requirements gathered in interaction with dance, sports and craft experts. This interaction happened partially in the scope of several European projects. These projects provided the necessary evaluation testbeds for the work described in this dissertation. The research focused on developing technologies for dance analysis. Several challenges were identified during these projects, including the need for automatic video content analysis and annotation of video footage. Addressing these challenges sets the research environment for the work described in this document. A new video annotation tool prototype, MotionNotes, was designed and developed to study integration possibilities. It incorporates intelligent techniques for movement analysis using video, employing automatic video annotation to efficiently identify and tag key actions or events within the footage. This situation represents a larger set of possibilities where machine learning can improve interaction and user experience in interactive tools. The prototype underwent an iterative development process characterized by con- tinuous improvement. User participation was instrumental to the evolution of this work, contributing significantly in each iteration to improve existing functionalities and introducing new features, such as using their own video content to discover movement patterns. After extensive evaluations, MotionNotes proved to be a success- ful instance of an intelligent video annotation tool, enabling techniques that make it possible to augment video information. Therefore, the insights and findings outlined in this document resulted from our exploration of interactive video annotation, com- bined with machine learning, as a representative of a larger class of systems where intelligent tools can help human tasks.
- Augmented Reality Serious Games Towards Introducing and Promoting Chemistry to PreteensPublication . Olim, Sandra Mónica Anjo Câmara; Nisi, Valentina; Romão, TeresaScience education, particularly physics, and chemistry, remains unpopular among students. Children’s perception of chemistry as a complex and strenuous subject starts at a young age, even before their first contact with the discipline, and these perceptions can affect the students’ achievements and future career choices in this domain. As a result, teachers and educational stakeholders acknowledge the need for innovative tools to reach the children and motivate them in these areas. This intervention should be applied to reach students earlier to demystify negative perceptions and encourage children to invest in these areas. Augmented Reality (AR) and Serious Games continue to be used as tools to engage students with learning content perceived as less engaging or challenging. However, in chemistry, the vast majority of AR educational tools are designed for older students (15 and older) and mainly focus on the potential of AR to provide the visualization of scientific phenomena. Our research revolves around AR multimodal Serious Games as an opportunity to explore different pedagogical methods to facilitate scientific content learning while incorporating game elements to engage preteen children (9 to 13 years old) with chemistry. This thesis systematically investigated the effectiveness of three Augmented Reality serious games designed around the chemistry learning experience. In the scope of this thesis, three novel AR serious games were created, namely: "Periodic Fable Discovery," "Periodic Fable An Augmented Journey," and "Periodic Fable in The Wild." These games were iterated, tested, and refined employing Research through Design methodology. The thesis contribution is centered around the design and lessons learned while producing three novel AR serious game artifacts. The outcomes of this research will establish the underpinnings for developing guidelines that can be implemented by game designers and developers who aim to create educational Augmented Reality (AR) Serious Games.
- Extended Skin: Designing Interactive Content for Ubiquitous Computing MaterialsPublication . Paiva, Isabel Maria da Silva; Câmara, AntónioCurrent research is inspired by the impact of digital media on disciplinary division. Sim- ultaneously, recognizes the difficulty of engineering (applied science) to consider the humani- ties as fundamental contributors in the process of making. Steaming from a design perspective, the intersection between art (design) and science, questions if these relations can open per- spectives on the matter of designing within a U.C. context, and fundamentally, introduces the question on how this can be done Furthermore, the motivation for this research arises from considering that innovation in technology is happening in the fields typically identified as engineering. And, despite this, the in-corporation of these inventions in life, considering some discussed exceptions, has not typ- ically been present in the concerns of design action and methods. Therefore, the challenge of current research is to contribute to the realm of ubiquitous computing, routed by design, to some degree aiming to contribute to the field. A deeper analysis into the subject of U.C., there is the realization that there is minority presence of the humanities in the discussion of U.C. (Dourish and Bell, 2011). Technological disruption offers continuous inspiration for design innovation within U.C. Furthermore, the inquiry labeled as “material turn” contextualizes a dialogue between nano- technology and traditional materials. Nanotechnology is applied to project development, while considering a human centred design approach. This focus is present throughout this disserta- tion. The research proposal describes SuberSkin, as a responsive surface that works as a screen. The exploration of aesthetical effects is focused on visual properties – using high con- trast between natural cork colors, dark and light brown. The proposal is highly experimental, and ultimately, aims to explore potential routes on cork research, linked to that of U.C. Thus, recreating and transforming this material into an intelligent surface. In sum, this thesis discusses displacement of disciplines suggested as having a positive impact in interdisciplinary thought and for future design. Therefore a methodology, "research through techne" is presented that illustrates this intention.
- Interactive Sound in Performance Ecologies: Studying Connections among Actors and ArtifactsPublication . Masu, Raul Giancarlo Maria; Correia, Nuno; Romão, TeresaThis thesis’s primary goal is to investigate performance ecologies, that is the compound of humans, artifacts and environmental elements that contribute to the result of a per- formance. In particular, this thesis focuses on designing new interactive technologies for sound and music. The goal of this thesis leads to the following Research Questions (RQs): • RQ1 How can the design of interactive sonic artifacts support a joint expression across different actors (composers, choreographers, and performers, musicians, and dancers) in a given performance ecology? • RQ2 How does each different actor influence the design of different artifacts, and what impact does this have on the overall artwork? • RQ3 How do the different actors in the same ecology interact, and appropriate an interactive artifact? To reply to these questions, a new framework named ARCAA has been created. In this framework, all the Actors of a given ecology are connected to all the Artifacts throughout three layers: Role, Context and Activity. This framework is then applied to one systematic literature review, two case studies on music performance and one case study in dance performance. The studies help to better understand the shaded roles of composers, per- formers, instrumentalists, dancers, and choreographers, which is relevant to better design interactive technologies for performances. Finally, this thesis proposes a new reflection on the blurred distinction between composing and designing a new instrument in a context that involves a multitude of actors. Overall, this work introduces the following contributions to the field of interaction design applied to music technology: 1) ARCAA, a framework to analyse the set of inter- connected relationship in interactive (music) performances, validated through 2 music studies, 1 dance study and 1 systematic literature analysis; 2) Recommendations for de- signing music interactive system for performance (music or dance), accounting for the needs of the various actors and for the overlapping on music composition and design of in- teractive technology; 3) A taxonomy of how scores have shaped performance ecologies in NIME, based on a systematic analysis of the literature on score in the NIME proceedings; 4) Proposal of a methodological approach combining autobiographical and idiographical design approaches in interactive performances.
- Cloud-edge hybrid applicationsPublication . Linde, Albert van der; Preguiça, Nuno; Leitão, JoãoMany modern applications are designed to provide interactions among users, including multi- user games, social networks and collaborative tools. Users expect application response time to be in the order of milliseconds, to foster interaction and interactivity. The design of these applications typically adopts a client-server model, where all interac- tions are mediated by a centralized component. This approach introduces availability and fault- tolerance issues, which can be mitigated by replicating the server component, and even relying on geo-replicated solutions in cloud computing infrastructures. Even in this case, the client-server communication model leads to unnecessary latency penalties for geographically close clients and high operational costs for the application provider. This dissertation proposes a cloud-edge hybrid model with secure and ecient propagation and consistency mechanisms. This model combines client-side replication and client-to-client propagation for providing low latency and minimizing the dependency on the server infras- tructure, fostering availability and fault tolerance. To realize this model, this works makes the following key contributions. First, the cloud-edge hybrid model is materialized by a system design where clients maintain replicas of the data and synchronize in a peer-to-peer fashion, and servers are used to assist clients’ operation. We study how to bring most of the application logic to the client-side, us- ing the centralized service primarily for durability, access control, discovery, and overcoming internetwork limitations. Second, we dene protocols for weakly consistent data replication, including a novel CRDT model (∆-CRDTs). We provide a study on partial replication, exploring the challenges and fundamental limitations in providing causal consistency, and the diculty in supporting client- side replicas due to their ephemeral nature. Third, we study how client misbehaviour can impact the guarantees of causal consistency. We propose new secure weak consistency models for insecure settings, and algorithms to enforce such consistency models. The experimental evaluation of our contributions have shown their specic benets and limitations compared with the state-of-the-art. In general, the cloud-edge hybrid model leads to faster application response times, lower client-to-client latency, higher system scalability as fewer clients need to connect to servers at the same time, the possibility to work oine or disconnected from the server, and reduced server bandwidth usage. In summary, we propose a hybrid of cloud-and-edge which provides lower user-to-user la- tency, availability under server disconnections, and improved server scalability – while being ecient, reliable, and secure.
- Leveraging eXtented Reality & Human-Computer Interaction for User Experi- ence in 360◦ VideoPublication . Bala, Paulo Alexandre Câmara; Nisi, Valentina; Correia, Nuno; Nunes, DuarteEXtended Reality systems have resurged as a medium for work and entertainment. While 360o video has been characterized as less immersive than computer-generated VR, its realism, ease of use and affordability mean it is in widespread commercial use. Based on the prevalence and potential of the 360o video format, this research is focused on improving and augmenting the user experience of watching 360o video. By leveraging knowledge from Extented Reality (XR) systems and Human-Computer Interaction (HCI), this research addresses two issues affecting user experience in 360o video: Attention Guidance and Visually Induced Motion Sickness (VIMS). This research work relies on the construction of multiple artifacts to answer the de- fined research questions: (1) IVRUX, a tool for analysis of immersive VR narrative expe- riences; (2) Cue Control, a tool for creation of spatial audio soundtracks for 360o video, as well as enabling the collection and analysis of captured metrics emerging from the user experience; and (3) VIMS mitigation pipeline, a linear sequence of modules (including optical flow and visual SLAM among others) that control parameters for visual modi- fications such as a restricted Field of View (FoV). These artifacts are accompanied by evaluation studies targeting the defined research questions. Through Cue Control, this research shows that non-diegetic music can be spatialized to act as orientation for users. A partial spatialization of music was deemed ineffective when used for orientation. Addi- tionally, our results also demonstrate that diegetic sounds are used for notification rather than orientation. Through VIMS mitigation pipeline, this research shows that dynamic restricted FoV is statistically significant in mitigating VIMS, while mantaining desired levels of Presence. Both Cue Control and the VIMS mitigation pipeline emerged from a Research through Design (RtD) approach, where the IVRUX artifact is the product of de- sign knowledge and gave direction to research. The research presented in this thesis is of interest to practitioners and researchers working on 360o video and helps delineate future directions in making 360o video a rich design space for interaction and narrative.
- Modeling Deception for Cyber SecurityPublication . Faveri, Cristiano De; Moreira, AnaIn the era of software-intensive, smart and connected systems, the growing power and so- phistication of cyber attacks poses increasing challenges to software security. The reactive posture of traditional security mechanisms, such as anti-virus and intrusion detection systems, has not been sufficient to combat a wide range of advanced persistent threats that currently jeopardize systems operation. To mitigate these extant threats, more ac- tive defensive approaches are necessary. Such approaches rely on the concept of actively hindering and deceiving attackers. Deceptive techniques allow for additional defense by thwarting attackers’ advances through the manipulation of their perceptions. Manipu- lation is achieved through the use of deceitful responses, feints, misdirection, and other falsehoods in a system. Of course, such deception mechanisms may result in side-effects that must be handled. Current methods for planning deception chiefly portray attempts to bridge military deception to cyber deception, providing only high-level instructions that largely ignore deception as part of the software security development life cycle. Con- sequently, little practical guidance is provided on how to engineering deception-based techniques for defense. This PhD thesis contributes with a systematic approach to specify and design cyber deception requirements, tactics, and strategies. This deception approach consists of (i) a multi-paradigm modeling for representing deception requirements, tac- tics, and strategies, (ii) a reference architecture to support the integration of deception strategies into system operation, and (iii) a method to guide engineers in deception mod- eling. A tool prototype, a case study, and an experimental evaluation show encouraging results for the application of the approach in practice. Finally, a conceptual coverage map- ping was developed to assess the expressivity of the deception modeling language created.
- CLASS: A Logical Foundation for Typeful Programming with Shared StatePublication . Rocha, Pedro Manuel Sabino; Caires, LuísSoftware construction depends on imperative state sharing and concurrency, which are naturally present in several application domains and are also exploited to improve the structure and efficiency of computer programs. However, reasoning about concurrency and shared mutable state is hard, error-prone and the source of many programming bugs, such as memory leaks, data corruption, deadlocks and non-termination. In this thesis, we develop CLASS: a core session-based language with a lightweight substructural type system, that results from a principled extension of the propositions-astypes correspondence with second-order classical linear logic. More concretely, CLASS offers support for session-based communication, mutex-protected first-class reference cells, dynamic state sharing, generic polymorphic algorithms, data abstraction and primitive recursion. CLASS expresses and types significant realistic programs, that manipulate memoryefficient linked data structures (linked lists, binary search trees) with support for updates in-place, shareable concurrent ADTs (counters, stacks, functional and imperative queues), resource synchronisation methods (fork-joins, barriers, dining philosophers, generic corecursive protocols). All of these examples are guaranteed to be safe, a result that follows by the logical approach. The linear logical foundations guarantee that well-typed CLASS programs do not go wrong: they never deadlock on communication or reference cell acquisition, do not leak memory and always terminate, even if they share complex data structures protected by synchronisation primitives. Furthermore, since we follow a propositions-as-types approach, we can reason about the behaviour of concurrent stateful processes by algebraic program manipulation. The feasibility of our approach is witnessed by the implementation of a type checker and interpreter for CLASS, which validates and guides the development of many realistic programs. The implementation is available with an open-source license, together with several examples.