Controllable Sequence Editing for Counterfactual Generation

2025-02-05

AI Summary

Existing counterfactual sequence generation methods lack fine-grained control over *when* and *where* edits take effect—particularly failing to support *localized delayed interventions*, i.e., edits triggered at a specified future timestep and affecting only a subset of co-occurring variables. To address this, we propose CLEF, the first controllable editing framework enabling *on-demand triggering*, *local activation*, and *single-shot generation at arbitrary future timesteps*. CLEF introduces a novel learnable temporal concept encoding mechanism, integrated with conditional intervention gating and local attention. Evaluated on cellular and patient trajectory data, CLEF reduces mean absolute error (MAE) by 36.01% for immediate edits and 65.71% for long-horizon counterfactual generation. A clinical case study on diabetes demonstrates that CLEF’s counterfactual interventions yield clinically plausible and actionable treatment recommendations.

Technology Category

• Machine Learning: Causal Learning
• Reasoning under Uncertainty: Causality
• Knowledge Representation and Reasoning: Action, Change, and Causality

Application Category

• Search and Retrieval-Augmented AI: Web learning to rank, online learning, and counterfactual learning for ranking
• User Modeling, Personalization and Recommendation: Attacks and countermeasures in recommendation systems
• Responsible Web: Machine-in-the-loop, human agency and autonomy

Abstract

Sequence models generate counterfactuals by modifying parts of a sequence based on a given condition, enabling reasoning about"what if"scenarios. While these models excel at conditional generation, they lack fine-grained control over when and where edits occur. Existing approaches either focus on univariate sequences or assume that interventions affect the entire sequence globally. However, many applications require precise, localized modifications, where interventions take effect only after a specified time and impact only a subset of co-occurring variables. We introduce CLEF, a controllable sequence editing model for counterfactual reasoning about both immediate and delayed effects. CLEF learns temporal concepts that encode how and when interventions should influence a sequence. With these concepts, CLEF selectively edits relevant time steps while preserving unaffected portions of the sequence. We evaluate CLEF on cellular and patient trajectory datasets, where gene regulation affects only certain genes at specific time steps, or medical interventions alter only a subset of lab measurements. CLEF improves immediate sequence editing by up to 36.01% in MAE compared to baselines. Unlike prior methods, CLEF enables one-step generation of counterfactual sequences at any future time step, outperforming baselines by up to 65.71% in MAE. A case study on patients with type 1 diabetes mellitus shows that CLEF identifies clinical interventions that shift patient trajectories toward healthier outcomes.

Problem

• Fine-grained control over sequence edits
• Localized modifications in multivariate sequences
• Counterfactual reasoning with temporal effects

Innovation

• Controllable sequence editing model
• Temporal concept learning
• Selective time step editing