Newest papers

Causal Integrated Gradients: Scalable Feature Attribution Along Causal Paths

Joran Michiels, Anis Ismail, Maarten De Vos, Alejandro Sifrim · 2026 · arXiv (Cornell University)

DOI: 10.48550/arxiv.2610.22126 · cited by 0 works (OpenAlex) · open-access PDF

French

Ces méthodes d'attribution de gradients fondées sur des cellules gratifiantes sont des méthodes d'attribution de cellules gratifiantes existantes qui comptent pour des dépendances de données de caractéristiques qui n'échelonnent pas à des dimensions élevées. Nous proposons des gradients transactionnels intégrés causaux (CIG) et des gradients anticipés causaux (CEG), qui étendent les gradients intégrés pour incorporer la structure causale préalable entre les caractéristiques d'entrée tout en conservant son efficacité computationnelle et, dans des conditions sur le graphique causal, ses garanties axiomatiques. Dans de nombreux paramètres, une structure causale entre les caractéristiques est déjà connue, par exemple, qu'un facteur de transcription régule causal

Machine translation (Meta M2M100); the English original follows.

Abstract

Existing feature attribution methods that account for feature dependencies do not scale to high dimensions. We propose causal integrated gradients (CIG) and causal expected gradients (CEG), which extend integrated gradients to incorporate prior causal structure among input features while retaining its computational efficiency and, under conditions on the causal graph, its axiomatic guarantees. In many settings, some causal structure between features is already known, for example, that a transcription factor causally regulates a target gene's expression. Yet existing gradient-based methods cannot exploit this structure, instead crediting a feature's downstream effects over its true upstream cause, such as ranking the target gene above the transcription factor that regulates it. Our key idea is to embed this causal structure by routing the integration path along the causal graph. The method thereby decomposes each feature's attribution into a direct effect and effects mediated through its causal descendants. We evaluate CIG in three settings of increasing realism. First, on a synthetic benchmark with a known causal model, CIG's attributions best match the ground-truth causal effect from Pearl's do-calculus, matching the accuracy of causal Shapley values at a fraction of the computational cost. Second, on simulated single-cell gene expression data with a known gene regulatory network (Dyngen), CIG most accurately recovers the transcription factors driving the simulation. Third, on real interferon-$β$ stimulation data, where only an approximate causal graph is available, CIG better prioritizes the transcription factors driving the measured response than DeepSHAP and integrated gradients. Together, these results show that incorporating causal structure into gradient-based attributions recovers upstream drivers that existing methods miss, without sacrificing scalability.

Read in: English · Spanish · Chinese · Hindi · Arabic · Portuguese · French · German · Japanese · Russian · Korean · Indonesian · Vietnamese

Topics: Computer science, Causal model, Causal structure, Attribution, Scalability, Artificial intelligence

Sources: OpenAlex (CC0), arXiv, Crossref. Agent Exchange