Communicating Credit Risk with Large Language Models: Evaluation of Explanations from Standard and Alternative Data-Based Models
研究利用大型语言模型将信用风险模型的复杂解释转化为易于理解的风险叙述，以解决现代信用风险模型解释过于技术化的问题。
0 citationsRead paper
Selected work
Representative Papers
A Topological Framework for Finite Behavioural Observations and Verification
This work investigates formal verification and runtime monitoring of properties in concurrent systems based on limited behavioral observations, such as traces or simulations. By introducing point-set topological methods, it establishes a precise correspondence between the topology induced by different observation mechanisms on the space of processes and verifiability: precisely those properties that are open sets in the respective topology are verifiable. The main contributions include a general verification theorem unifying monitorability under trace, simulation, and finite-depth bisimulation semantics; a rigorous proof that the topologies τ_O and τ_sim, induced respectively by observational closure and simulation relations, satisfy a strict inclusion τ_sim ⊂ τ_O; and the insight that stronger behavioral equivalences yield fundamentally distinct topologies, thereby deepening the understanding of the relationship between behavioral semantics and verification capabilities.
0 citationsRead paper
Does Synthetic Data Help? Empirical Evidence from Deep Learning Time Series Forecasters
This study systematically investigates the effectiveness of synthetic data in time series forecasting and its dependence on model architecture. Drawing on 4,218 experiments across nine configurations, the authors evaluate five prominent deep learning models—including TimesNet, iTransformer, DLinear, and PatchTST—on four synthetic signals and seven real-world datasets. The work reveals, for the first time, that the benefits of synthetic data are highly architecture-dependent: channel-mixing models consistently gain substantial improvements, particularly under low-data regimes, whereas synthetic data proves detrimental in 67% of all experimental settings. The study proposes effective usage strategies tailored to channel-mixing architectures and demonstrates that progressive scheduling outperforms hard curriculum switching. Among synthetic generation methods, only the seasonal-trend decomposition generator yields consistent performance gains.
0 citationsRead paper
An Undecidability Proof for the Plan Existence Problem
This study addresses the planning existence problem in epistemic reasoning: given a modal-logic goal, an initial epistemic state, and a set of epistemic actions, does there exist a sequence of actions that achieves the goal? Focusing on a highly restricted setting where epistemic actions have preconditions of modal depth at most one and no postconditions, the analysis employs modal logic, Kripke semantics, and formal reduction techniques. The work establishes, for the first time, that even under these strong syntactic and semantic constraints, the planning existence problem remains undecidable. This result fills a critical gap in the decidability landscape of epistemic planning and delineates new theoretical boundaries for automated planning and cognitive reasoning systems.
0 citationsRead paper
Can SOC Operators Explain their Decisions while Triaging Alarms? A Real-World Study
This study addresses a critical gap in explainability within Security Operations Center (SOC) workflows: while SOC analysts achieve high decision accuracy (83%) during alert triage, their post-hoc explanations align with the true root causes in only 39% of cases. Through a systematic literature review encompassing 257 papers and an empirical user study involving 12 SOC analysts, this work provides the first evidence of a significant disconnect between decision correctness and explanation fidelity. The findings underscore the urgent need for computational mechanisms that support accurate, causally grounded justifications for analyst decisions. By revealing this explanatory deficit, the research offers foundational insights for enhancing human–machine collaboration and bolstering trustworthiness in SOC decision-making processes.
0 citationsRead paper
Recent publications
Latest Papers
Communicating Credit Risk with Large Language Models: Evaluation of Explanations from Standard and Alternative Data-Based Models
研究利用大型语言模型将信用风险模型的复杂解释转化为易于理解的风险叙述，以解决现代信用风险模型解释过于技术化的问题。
0 citationsRead paper
A Topological Framework for Finite Behavioural Observations and Verification
This work investigates formal verification and runtime monitoring of properties in concurrent systems based on limited behavioral observations, such as traces or simulations. By introducing point-set topological methods, it establishes a precise correspondence between the topology induced by different observation mechanisms on the space of processes and verifiability: precisely those properties that are open sets in the respective topology are verifiable. The main contributions include a general verification theorem unifying monitorability under trace, simulation, and finite-depth bisimulation semantics; a rigorous proof that the topologies τ_O and τ_sim, induced respectively by observational closure and simulation relations, satisfy a strict inclusion τ_sim ⊂ τ_O; and the insight that stronger behavioral equivalences yield fundamentally distinct topologies, thereby deepening the understanding of the relationship between behavioral semantics and verification capabilities.
0 citationsRead paper
Does Synthetic Data Help? Empirical Evidence from Deep Learning Time Series Forecasters
This study systematically investigates the effectiveness of synthetic data in time series forecasting and its dependence on model architecture. Drawing on 4,218 experiments across nine configurations, the authors evaluate five prominent deep learning models—including TimesNet, iTransformer, DLinear, and PatchTST—on four synthetic signals and seven real-world datasets. The work reveals, for the first time, that the benefits of synthetic data are highly architecture-dependent: channel-mixing models consistently gain substantial improvements, particularly under low-data regimes, whereas synthetic data proves detrimental in 67% of all experimental settings. The study proposes effective usage strategies tailored to channel-mixing architectures and demonstrates that progressive scheduling outperforms hard curriculum switching. Among synthetic generation methods, only the seasonal-trend decomposition generator yields consistent performance gains.
0 citationsRead paper
An Undecidability Proof for the Plan Existence Problem
This study addresses the planning existence problem in epistemic reasoning: given a modal-logic goal, an initial epistemic state, and a set of epistemic actions, does there exist a sequence of actions that achieves the goal? Focusing on a highly restricted setting where epistemic actions have preconditions of modal depth at most one and no postconditions, the analysis employs modal logic, Kripke semantics, and formal reduction techniques. The work establishes, for the first time, that even under these strong syntactic and semantic constraints, the planning existence problem remains undecidable. This result fills a critical gap in the decidability landscape of epistemic planning and delineates new theoretical boundaries for automated planning and cognitive reasoning systems.
0 citationsRead paper
Can SOC Operators Explain their Decisions while Triaging Alarms? A Real-World Study
This study addresses a critical gap in explainability within Security Operations Center (SOC) workflows: while SOC analysts achieve high decision accuracy (83%) during alert triage, their post-hoc explanations align with the true root causes in only 39% of cases. Through a systematic literature review encompassing 257 papers and an empirical user study involving 12 SOC analysts, this work provides the first evidence of a significant disconnect between decision correctness and explanation fidelity. The findings underscore the urgent need for computational mechanisms that support accurate, causally grounded justifications for analyst decisions. By revealing this explanatory deficit, the research offers foundational insights for enhancing human–machine collaboration and bolstering trustworthiness in SOC decision-making processes.
0 citationsRead paper