Clemens Dubslaff

Christel Baier, Clemens Dubslaff, Patrick Wienhöft et al.

A central task in control theory, artificial intelligence, and formal methods is to synthesize reward-maximizing strategies for agents that operate in partially unknown environments. In environments modeled by gray-box Markov decision processes (MDPs), the impact of the agents' actions are known in terms of successor states but not the stochastics involved. In this paper, we devise a strategy synthesis algorithm for gray-box MDPs via reinforcement learning that utilizes interval MDPs as internal model. To compete with limited sampling access in reinforcement learning, we incorporate two novel concepts into our algorithm, focusing on rapid and successful learning rather than on stochastic guarantees and optimality: lower confidence bound exploration reinforces variants of already learned practical strategies and action scoping reduces the learning action space to promising actions. We illustrate benefits of our algorithms by means of a prototypical implementation applied on examples from the AI and formal methods communities.

Christel Baier, Clemens Dubslaff, Holger Hermanns et al.

Bayesian networks (BNs) are a probabilistic graphical model widely used for representing expert knowledge and reasoning under uncertainty. Traditionally, they are based on directed acyclic graphs that capture dependencies between random variables. However, directed cycles can naturally arise when cross-dependencies between random variables exist, e.g., for modeling feedback loops. Existing methods to deal with such cross-dependencies usually rely on reductions to BNs without cycles. These approaches are fragile to generalize, since their justifications are intermingled with additional knowledge about the application context. In this paper, we present a foundational study regarding semantics for cyclic BNs that are generic and conservatively extend the cycle-free setting. First, we propose constraint-based semantics that specify requirements for full joint distributions over a BN to be consistent with the local conditional probabilities and independencies. Second, two kinds of limit semantics that formalize infinite unfolding approaches are introduced and shown to be computable by a Markov chain construction.

Debraj Chakraborty, Clemens Dubslaff, Sudeep Kanav et al.

Safety-critical controllers of complex systems are hard to construct manually. Automated approaches such as controller synthesis or learning provide a tempting alternative but usually lack explainability. To this end, learning decision trees (DTs) have been prevalently used towards an interpretable model of the generated controllers. However, DTs do not exploit shared decision-making, a key concept exploited in binary decision diagrams (BDDs) to reduce their size and thus improve explainability. In this work, we introduce predicate decision diagrams (PDDs) that extend BDDs with predicates and thus unite the advantages of DTs and BDDs for controller representation. We establish a synthesis pipeline for efficient construction of PDDs from DTs representing controllers, exploiting reduction techniques for BDDs also for PDDs.

Patrick Wienhöft, Marnix Suilen, Thiago D. Simão et al.

In an offline reinforcement learning setting, the safe policy improvement (SPI) problem aims to improve the performance of a behavior policy according to which sample data has been generated. State-of-the-art approaches to SPI require a high number of samples to provide practical probabilistic guarantees on the improved policy's performance. We present a novel approach to the SPI problem that provides the means to require less data for such guarantees. Specifically, to prove the correctness of these guarantees, we devise implicit transformations on the data set and the underlying environment model that serve as theoretical foundations to derive tighter improvement bounds for SPI. Our empirical evaluation, using the well-established SPI with baseline bootstrapping (SPIBB) algorithm, on standard benchmarks shows that our method indeed significantly reduces the sample complexity of the SPIBB algorithm.

Clemens Dubslaff, Kallistos Weis, Christel Baier et al.

Detecting and understanding reasons for defects and inadvertent behavior in software is challenging due to their increasing complexity. In configurable software systems, the combinatorics that arises from the multitude of features a user might select from adds a further layer of complexity. We introduce the notion of feature causality, which is based on counterfactual reasoning and inspired by the seminal definition of actual causality by Halpern and Pearl. Feature causality operates at the level of system configurations and is capable of identifying features and their interactions that are the reason for emerging functional and non-functional properties. We present various methods to explicate these reasons, in particular well-established notions of responsibility and blame that we extend to the feature-oriented setting. Establishing a close connection of feature causality to prime implicants, we provide algorithms to effectively compute feature causes and causal explications. By means of an evaluation on a wide range of configurable software systems, including community benchmarks and real-world systems, we demonstrate the feasibility of our approach: We illustrate how our notion of causality facilitates to identify root causes, estimate the effects of features, and detect feature interactions.

Clemens Dubslaff, Sascha Klüppelholz, Christel Baier

In a software product line (SPL), a collection of software products is defined by their commonalities in terms of features rather than explicitly specifying all products one-by-one. Several verification techniques were adapted to establish temporal properties of SPLs. Symbolic and family-based model checking have been proven to be successful for tackling the combinatorial blow-up arising when reasoning about several feature combinations. However, most formal verification approaches for SPLs presented in the literature focus on the static SPLs, where the features of a product are fixed and cannot be changed during runtime. This is in contrast to dynamic SPLs, allowing to adapt feature combinations of a product dynamically after deployment. The main contribution of the paper is a compositional modeling framework for dynamic SPLs, which supports probabilistic and nondeterministic choices and allows for quantitative analysis. We specify the feature changes during runtime within an automata-based coordination component, enabling to reason over strategies how to trigger dynamic feature changes for optimizing various quantitative objectives, e.g., energy or monetary costs and reliability. For our framework there is a natural and conceptually simple translation into the input language of the prominent probabilistic model checker PRISM. This facilitates the application of PRISM's powerful symbolic engine to the operational behavior of dynamic SPLs and their family-based analysis against various quantitative queries. We demonstrate feasibility of our approach by a case study issuing an energy-aware bonding network device.