PLOT: Progressive Localization via Optimal Transport in Neural Causal Abstraction

Abstract

Causal abstraction offers a principled framework for mechanistic interpretability, aligning a high-level causal model with low-level neural computation through interchange intervention analysis. Finding such an alignment, however, often requires fitting and evaluating separate learned mappings across many candidate neural locations. We introduce PLOT, a gradient-free approach for joint correspondence discovery via a global matching of intervention effects. PLOT represents abstract and neural interventions by geometric signatures of their effects on the shared task output and fits an optimal transport coupling between the two collections. The coupling can be calibrated directly into an executable intervention handle or used progressively to select coarse parent sites for finer localization, reducing the cost of searching broad collections of neural sites. Experiments on hierarchical equality, binary addition, and multiple-choice question answering (MCQA) demonstrate fast and accurate direct handles without learning intervention rotations and show that progressive localization can improve accuracy under intervention-size constraints while reducing computational cost. PLOT can also guide gradient-based subspace learning, with PLOT-guided distributed alignment search (DAS) attaining accuracy comparable to full DAS at approximately 3.6× and 20× lower serial runtime on binary addition and MCQA, respectively. PLOT thus separates the search for candidate correspondences from optional subspace learning and provides a common framework for constructing and testing neural intervention handles for a given causal model.

Relevance Propagation as Entropic Optimal Transport on Filippov Neural Networks: Sinkhorn Propagation and Region Relevance in Image Classifiers
Olivier Gibaru, Eric Nyiri
arXiv3 Nov 2026PreprintCC BY 4.0Full textOptimisation & controlarXiv
Optimal Watermark Localization in Mixed-Source Large Language Model Texts
Jose H. Blanchet, T. Tony Cai, Xiang Li +3
arXiv15 Oct 2026PreprintCC BY 4.0Full textStatisticsarXiv
Stable q-Hermitian coordinate calculus: Capelli--PBW transport, mixed polarization algebra, and localization obstructions
Baruch Schneider, Diana Schneiderová, Yifan Zhang
arXiv24 Sep 2026PreprintCC BY-NC-ND 4.0Full textAnalysis & PDEsarXiv
Convergence of Coordinate Ascent Variational Inference for log-concave measures via optimal transport
Manuel Arnese, Daniel Lacker
The Annals of Applied Probability1 Jun 2026Peer-reviewedLicence not statedFull text5 citedStatisticsOpenAlex
Hausdorff dimension of the singular set in optimal transport
Shibing Chen, Qi-Rui Li, Yuanyuan Li +2
arXiv20 Oct 2026PreprintCC BY 4.0Full textAnalysis & PDEsarXiv

Cite

BibTeX RIS

BibTeX

@misc{chang2026plot,
  title = {{PLOT: Progressive Localization via Optimal Transport in Neural Causal Abstraction}},
  author = {Jonathn Chang and Arya Datla and Ziv Goldfeld},
  howpublished = {arXiv},
  year = {2026},
  month = oct,
  doi = {10.48550/arxiv.2605.06979},
  eprint = {2605.06979},
  archivePrefix = {arXiv},
  url = {https://arxiv.org/abs/2605.06979},
}

RIS

TY  - UNPB
TI  - PLOT: Progressive Localization via Optimal Transport in Neural Causal Abstraction
AU  - Jonathn Chang
AU  - Arya Datla
AU  - Ziv Goldfeld
PY  - 2026
DA  - 2026/10/04
PB  - arXiv
DO  - 10.48550/arxiv.2605.06979
UR  - https://arxiv.org/abs/2605.06979
AB  - Causal abstraction offers a principled framework for mechanistic interpretability, aligning a high-level causal model with low-level neural computation through interchange intervention analysis. Finding such an alignment, however, often requires fitting and evaluating separate learned mappings across many candidate neural locations. We introduce PLOT, a gradient-free approach for joint correspondence discovery via a global matching of intervention effects. PLOT represents abstract and neural interventions by geometric signatures of their effects on the shared task output and fits an optimal transport coupling between the two collections. The coupling can be calibrated directly into an executable intervention handle or used progressively to select coarse parent sites for finer localization, reducing the cost of searching broad collections of neural sites. Experiments on hierarchical equality, binary addition, and multiple-choice question answering (MCQA) demonstrate fast and accurate direct handles without learning intervention rotations and show that progressive localization can improve accuracy under intervention-size constraints while reducing computational cost. PLOT can also guide gradient-based subspace learning, with PLOT-guided distributed alignment search (DAS) attaining accuracy comparable to full DAS at approximately 3.6× and 20× lower serial runtime on binary addition and MCQA, respectively. PLOT thus separates the search for candidate correspondences from optional subspace learning and provides a common framework for constructing and testing neural intervention handles for a given causal model.
ER  -

Abstract

Related

Cite