Tag
This paper introduces causal retention in interactive agents, examining how frozen learned states can answer intervention queries independently of training objectives, supported by theoretical analysis and experiments on finite causal systems and AI models like Qwen.
The paper proposes bipartite graphical causal models (BGCMs) to resolve ambiguities in causal interventions for systems at equilibrium with cyclic dependencies, generalizing existing frameworks like causal Bayesian networks and structural causal models.
Introduces Adversarial Causal Intervention Falsification (ACIF), a sequential game where a structural causal generator proposes observational and interventional distributions while an adversarial experimentalist selects interventions to falsify it. The paper provides theoretical guarantees, including finite-sample convergence and model-selection, bridging causal generative modeling, active discovery, and experimental design.
This paper extends Pearl's structural causal model framework by introducing causal zeros and causal differential equations to handle symmetric constraints and feedback cycles, which are not allowed in directed acyclic graphs.
This paper introduces relational structural causal models, extending structural causal models to settings with varying objects and relations. It provides theoretical results for identification and proposes relational neural causal models that outperform non-relational baselines on simulated traffic scenes.
This paper introduces COAST, a causal-intelligence framework for designing constraint-aware interventions that drive complex systems between states, integrating causal discovery, modeling, and multi-objective optimization to identify minimal effective interventions with mechanistic rationales.
CausaLab is a scalable environment for evaluating LLM agents on interactive causal discovery, assessing both predictive accuracy and faithful recovery of underlying causal mechanisms. Experiments reveal a gap between prediction and mechanism recovery, highlighting limits in current LLM agents as experimental causal reasoners.
This article introduces ReplaySCM, a benchmark designed to evaluate language models' ability to induce executable causal mechanisms from interventional evidence, focusing on semantic replay behavior rather than syntactic matches.