IT Faculty / IT-fakultetenhttps://hdl.handle.net/2077/99072024-03-28T15:16:08Z2024-03-28T15:16:08ZUnraveling The Black Box - Building Understandable AI Through Strategic Explanation and User-based DesignYu, Shurenhttps://hdl.handle.net/2077/804022024-03-13T21:03:14Z2024-01-01T00:00:00ZUnraveling The Black Box - Building Understandable AI Through Strategic Explanation and User-based Design
Yu, Shuren
The pervasive integration of Artificial Intelligence (AI) in society presents
both opportunities and challenges, with the black-box issue emerging as a
significant obstacle in realizing the full potential of AI. The opaque nature of
AI decision-making processes impedes user understanding, particularly
among non-technical individuals, raising concerns about the reliability of AI
recommendations. Therefore, how to help users understand AI decisionmaking
has become an urgent task. This thesis aims to assist developers in
contemplating how to construct AI that users can understand. To build
understandable AI, researchers have proposed many theories, methods, and
frameworks in existing research. However, there are still limitations and
challenges in current research. To address these challenges and finish the
research aim, starting with a discussion on transparency and interpretability,
the thesis elaborates on how to strategically explain to users within three
dimensions: simplifying algorithm, appropriate information disclosure, and
high-level collaboration. Furthermore, the thesis conducts surveys on users in
four high-stakes areas, establishing AI explainability principles based on
three stages, conceptualization, construction, and measurement. In addition to
these primary contributions, the thesis also covers some supportive work,
including challenges faced by explainable AI, user-centered development,
and automation trust. These works lay a solid foundation for addressing
research questions and achieving research objectives, while also providing
room for contemplation in future research.
2024-01-01T00:00:00ZFair Omega-regular GamesHausmann, DanielPiterman, NirSaglam, IrmakSchmuck, Anne-Kathrinhttps://hdl.handle.net/2077/801092024-02-27T21:11:23Z2024-01-01T00:00:00ZFair Omega-regular Games
Hausmann, Daniel; Piterman, Nir; Saglam, Irmak; Schmuck, Anne-Kathrin
We consider two-player games over finite graphs in which both players are restricted by fairness constraints on their moves. Given a two player game graph G=(V,E) and a set of fair moves E_f a subset of E a player is said to play fair in G if they choose an edge e in E_f infinitely often whenever the source vertex of e is visited infinitely often. Otherwise, they play unfair. We equip such games with two omega-regular winning conditions alpha and beta deciding the winner of mutually fair and mutually unfair plays, respectively. Whenever one player plays fair and the other plays unfair, the fairly playing player wins the game. The resulting games are called fair alpha/beta games. We formalize fair alpha/beta games and show that they are determined. For fair parity/parity games, i.e., fair alpha/beta games where alpha and beta are given each by a parity condition over G, we provide a polynomial reduction to (normal) parity games via a gadget construction inspired by the reduction of stochastic parity games to parity games. We further give a direct symbolic fixpoint algorithm to solve fair parity/parity games. On a conceptual level, we illustrate the translation between the gadget-based reduction and the direct symbolic algorithm which uncovers the underlying similarities of solution algorithms for fair and stochastic parity games, as well as for the recently considered class of fair games in which only one player is restricted by fair moves.
2024-01-01T00:00:00ZSymbolic Solution of Emerson-Lei Games for Reactive SynthesisHausmann, DanielLehaut, MathieuPiterman, Nirhttps://hdl.handle.net/2077/801082024-02-27T21:02:55Z2024-01-01T00:00:00ZSymbolic Solution of Emerson-Lei Games for Reactive Synthesis
Hausmann, Daniel; Lehaut, Mathieu; Piterman, Nir
Emerson-Lei conditions have recently attracted attention due to both their succinctness and their favorable closure properties. In the current work, we show how infinite-duration games with Emerson-Lei objectives can be analyzed in two different ways. First, we show that the Zielonka tree of the Emerson-Lei condition naturally gives rise to a new reduction to parity games. This reduction, however, does not result in optimal analysis. Second, we show based on the first reduction (and the Zielonka tree) how to provide a direct fixpoint-based characterization of the winning region. The fixpoint-based characterization allows for symbolic analysis. It generalizes the solutions of games with known winning conditions such as B\"uchi, GR[1], parity, Streett, Rabin and Muller objectives, and in the case of these conditions reproduces previously known symbolic algorithms and complexity results. We also show how the capabilities of the proposed algorithm can be exploited in reactive synthesis, suggesting a new expressive fragment of LTL that can be handled symbolically. Our fragment combines a safety specification and a liveness part. The safety part is unrestricted and the liveness part allows to define Emerson-Lei conditions on occurrences of letters. The symbolic treatment is enabled due to the simplicity of determinization in the case of safety languages and by using our new algorithm for game solving. This approach maximizes the number of steps solved symbolically in order to maximize the potential for efficient symbolic implementations.
2024-01-01T00:00:00ZPhenomenal Fields Forever. - Instructed Action and Perception’s WorkIvarsson, Jonashttps://hdl.handle.net/2077/794302023-12-20T21:06:15Z2023-01-01T00:00:00ZPhenomenal Fields Forever. - Instructed Action and Perception’s Work
Ivarsson, Jonas
Falkenberg, Mårten
2023-01-01T00:00:00Z