mlsanityData to 13 Oct 2026

Exploratory Causal Inference in SAEnce

arXiv 2510.14073 ↗First posted 15 Oct 2025cs.LG cs.AIcited by 4

Not on the map. Why?

Abstract (arXiv)

Randomized Controlled Trials are one of the pillars of science; nevertheless, they rely on hand-crafted hypotheses and expensive analysis. Such constraints prevent causal effect estimation at scale, potentially anchoring on popular yet incomplete hypotheses. We propose to discover the unknown effects of a treatment directly from data. For this, we turn unstructured data from a trial into meaningful representations via pretrained foundation models and interpret them via a sparse autoencoder. However, discovering significant causal effects at the neural level is not trivial due to multiple-testing issues and effects entanglement. To address these challenges, we introduce Neural Effect Search, a novel recursive procedure solving both issues by progressive stratification. After assessing the robustness of our algorithm on semi-synthetic experiments, we showcase, in the context of experimental ecology, the first successful unsupervised causal effect identification on a real-world scientific trial.

Key facts. Cited by 4 papers in the map (data through 13 Oct 2026). Builds on 32 of its 55 references that are in the map. First posted on arXiv 15 Oct 2025.

Builds on

32 of 55 references.

References · 32

  1. 2023DINOv2: Learning Robust Visual Features without Supervisioncited by 7,857
  2. 2023Sigmoid Loss for Language Image Pre-Trainingcited by 3,068
  3. 2023Sparse Autoencoders Find Highly Interpretable Features in Language Modelscited by 1,391
  4. 2022Toy Models of Superpositioncited by 950
  5. 2018Challenging Common Assumptions in the Unsupervised Learning of Disentangled Representationscited by 934
  6. 2023The Linear Representation Hypothesis and the Geometry of Large Language Modelscited by 706
  7. 2021Deep ViT Features as Dense Visual Descriptorscited by 329
  8. 2016Generalized Random Forestscited by 310
  9. 2015Recursive Partitioning for Heterogeneous Causal Effectscited by 255
  10. 2017Emergence of Invariance and Disentanglement in Deep Representationscited by 213
  11. 2025AxBench: Steering LLMs? Even Simple Baselines Outperform Sparse Autoencoderscited by 209
  12. 2024A is for Absorption: Studying Feature Splitting and Absorption in Sparse Autoencoderscited by 156
20 more
  1. 2017Causal Consistency of Structural Equation Modelscited by 116
  2. 2023The geometry of hidden representations of large transformer modelscited by 109
  3. 2025Sparse Autoencoders Do Not Find Canonical Units of Analysiscited by 85
  4. 2023Identifiability Guarantees for Causal Disentanglement from Soft Interventionscited by 84
  5. 2025Projecting Assumptions: The Duality Between Sparse Autoencoders and Concept Geometrycited by 60
  6. 2014Visual Causal Feature Learningcited by 59
  7. 2023Score-based Causal Representation Learning with Interventionscited by 49
  8. 2023RxRx1: A Dataset for Evaluating Experimental Batch Correction Methodscited by 45
  9. 2023General Identifiability and Achievability for Causal Representation Learningcited by 34
  10. 2023Leveraging sparse and shared feature activations for disentangled representation learningcited by 32
  11. 2025Interpreting the linear structure of vision-language model embedding spacescited by 32
  12. 2024Unifying Causal Representation Learning with the Invariance Principlecited by 31
  13. 2025Automated Interpretability Metrics Do Not Distinguish Trained and Random Transformerscited by 30
  14. 2025Sparse Autoencoders for Hypothesis Generationcited by 26
  15. 2025Position: Use Sparse Autoencoders to Discover Unknownscited by 18
  16. 2020Causal Rule Ensemble: Interpretable Discovery and Inference of Heterogeneous Treatment Effectscited by 13
  17. 2024Smoke and Mirrors in Causal Downstream Taskscited by 12
  18. 2025On the Limits of Sparse Autoencoders: A Theoretical Framework and Reweighted Remedycited by 12
  19. 2025Prediction-Powered Causal Inferencescited by 9
  20. 2025Binary Sparse Coding for Interpretabilitycited by 3

Built on by

Cited by year (2026 to 13 Oct)

Citing papers per year: 2026: 4.

4 papers.

  1. 2026The Observer Effect in World Models: Invasive Adaptation Corrupts Latent Physicscited by 3
  2. 2026Toward Identifiable Sparse Autoencoderscited by 2
  3. 2026Reconciling Causality and Non-Equilibrium Thermodynamics with Hamiltonian Causal Modelscited by 0
  4. 2026From Tokens to Policy: Causal and Interpretable Heterogeneous Treatment Effects Identificationcited by 0

Similar, not cited

Close in wording, no citation either way

  1. 2026Exploring Sparse Autoencoders in Text-Based Causal Confounding Adjustment
  2. 2026Causal Effect Estimation with Latent Textual Treatments
  3. 2025The Third Pillar of Causal Analysis? A Measurement Perspective on Causal Representations
  4. 2023Towards Causal Foundation Model: on Duality between Causal Inference and Attention
  5. 2024Deep Causal Inference for Point-referenced Spatial Data with Continuous Treatments
  6. 2026From Local to Cluster: A Unified Framework for Causal Discovery with Latent Variables
  7. 2026CausalArena: Benchmarking Causal Discovery in the Foundation Model Era
  8. 2025Addressing divergent representations from causal interventions on neural networks