Institution profile

Max Planck Institute for Software Systems

Academic institutioneurope · de
Official website
Research library251linked papers
Opportunities0open roles
Selected work

Representative Papers

Reinforcement Learning with Stochastic Reward Machines

Jun 28, 2022AAAI Conference on Artificial Intelligence

Existing reward machines (RMs) assume noise-free reward signals, limiting their applicability in real-world settings characterized by sparse, action-sequence-dependent, and noisy rewards. Method: This paper proposes the Stochastic Reward Machine (SRM), a novel RM variant that explicitly models stochastic reward observations. We introduce constraint solving into RM learning for the first time, enabling automatic inference of state partitions and transition relations from agent exploration trajectories to synthesize a minimal SRM. Theoretical analysis establishes asymptotic convergence to an optimal policy under reward noise. Results: Experiments on two representative noisy-reward tasks demonstrate that our approach significantly outperforms existing RM-based methods and naive denoising baselines, validating its robustness and effectiveness in learning from unreliable reward signals.

30 citations1 influentialRead paper

On the Decidability of Monadic Second-Order Logic with Arithmetic Predicates

May 13, 2024Logic in Computer Science

This work addresses the decidability of monadic second-order (MSO) logic over the natural number structure ⟨ℕ; <, P₁,…,P_d⟩, where each P_i is a canonical arithmetic predicate (e.g., powers of k, perfect k-th powers ℕ^k, or the Fibonacci sequence Fib). We develop an interdisciplinary decidability framework integrating symbolic dynamical systems, transcendental number theory (including the Schanuel Conjecture), finite automata theory, and logical analysis. Unconditionally—i.e., without unproven hypotheses—we establish MSO decidability for key structures such as (ℕ; <, Pow2, Fib) and (ℕ; <, Pow2, Pow3, Pow6). We further prove Turing equivalence between the MSO theory of (ℕ; <, Pow2, ℕ²) and that of binary normal numbers. Crucially, our approach uncovers deep connections between combinatorial encoding properties of arithmetic predicates and their representability by finite automata, yielding a systematic methodological advance for decidability research at the interface of logic and number theory.

7 citations1 influentialRead paper

Root Cause Analysis of Outliers with Missing Structural Knowledge

Jun 07, 2024arXiv.org

Real-world root cause analysis (RCA) faces a critical challenge: post-intervention distributions often contain only a few—or even a single—sample, rendering distribution-dependent or low-density-region regression methods statistically ill-posed. This paper proposes a lightweight root cause identification framework that requires neither counterfactual reasoning nor a fully specified structural causal model (SCM). It operates either given a causal DAG or, in the absence of one, solely from an anomaly score ranking. We theoretically prove that low-scoring anomalies rarely trigger high-scoring ones and derive a probabilistic upper bound on non-monotonic propagation paths. By abandoning Shapley-value-based attribution and density-sensitive regression, our method achieves linear time complexity O(n). It eliminates SCM fitting and counterfactual computation while providing rigorous theoretical guarantees and strong empirical performance.

4 citations1 influentialRead paper

Length Generalization Bounds for Transformers

Feb 13, 2026

This study investigates the length generalization capability of Transformers—specifically, their ability to generalize to arbitrarily long inputs when trained on sequences of bounded length—and the computability of associated generalization bounds. Leveraging formal language theory and computational complexity analysis, and employing C-RASP, a mathematically rigorous abstraction of the Transformer architecture, the work establishes for the first time that C-RASP models with two or more layers admit no computable length generalization bound. In contrast, both the positive fragment of C-RASP and fixed-precision Transformers possess tight, optimal, and computable exponential generalization bounds. These results uncover fundamental limitations inherent in deep Transformer models regarding length generalization while providing an exact characterization of the conditions under which such bounds remain computable.

1 citationsRead paper

The Algorithmic Self-Portrait: Deconstructing Memory in ChatGPT

Feb 01, 2026

This study addresses the opacity of memory mechanisms in conversational AI systems, which poses significant risks to user privacy, agency, and profile accuracy. Drawing on a novel dataset of 2,050 memory entries from 80 real ChatGPT users, the research employs content analysis, GDPR-based personal data classification, psychological inference detection, and query rewriting techniques to empirically demonstrate that 96% of these memories are unilaterally generated by the system. Among them, 28% contain personal data as defined under the GDPR, and 52% encode psychological insights. To mitigate these concerns, the work proposes Attribution Shield—a user empowerment framework that proactively alerts users to sensitive inferences and recommends query rewrites—thereby significantly enhancing user control over AI-generated memories while preserving interactive utility.

1 citationsRead paper
Recent publications

Latest Papers

Learning Metastable Dynamics

Sep 13, 2026

为解决物理系统中识别和分析亚稳态现象的挑战,提出了一种基于Koopman理论的新框架,通过学习系统的动态表示来提前预测亚稳态行为。

0 citationsRead paper