mlsanityData to 7 Oct 2026

CIRCUS: Circuit Consensus under Uncertainty via Stability Ensembles

arXiv 2603.00523 ↗First posted 28 Feb 2026cs.CL cs.AI cs.LGcited by 1

Not on the map. Why?

Abstract (arXiv)

Every mechanistic circuit carries an invisible asterisk: it reflects not just the model's computation, but the analyst's choice of pruning threshold. Change that choice and the circuit changes, yet current practice treats a single pruned subgraph as ground truth with no way to distinguish robust structure from threshold artifacts. We introduce CIRCUS, which reframes circuit discovery as a problem of uncertainty over explanations. CIRCUS prunes one attribution graph under B configurations, assigns each edge an empirical inclusion frequency s(e) in [0,1] measuring how robustly it survives across the configuration family, and extracts a consensus circuit of edges present in every view. This yields a principled core/contingent/noise decomposition (analogous to posterior model-inclusion indicators in Bayesian variable selection) that separates robust structure from threshold-sensitive artifacts, with negligible overhead. On Gemma-2-2B and Llama-3.2-1B, consensus circuits are 40x smaller than the union of all configurations while retaining comparable influence-flow explanatory power, consistently outperform influence-ranked and random baselines, and are confirmed causally relevant by activation patching.

Key facts. Cited by 1 paper in the map (data through 7 Oct 2026). Builds on 14 of its 19 references that are in the map. First posted on arXiv 28 Feb 2026.

Builds on

14 of 19 references.

References · 14

  1. 2017Towards A Rigorous Science of Interpretable Machine Learningcited by 1,400
  2. 2023Towards Best Practices of Activation Patching in Language Models: Metrics and Methodscited by 328
  3. 2025Open Problems in Mechanistic Interpretabilitycited by 173
  4. 2024A Practical Review of Mechanistic Interpretability for Transformer-Based Language Modelscited by 113
  5. 2022Rethinking Stability for Attribution-based Explanationscited by 52
  6. 2024Explaining Predictive Uncertainty by Exposing Second-Order Effectscited by 9
  7. 2020My Teacher Thinks The World Is Flat! Interpreting Automatic Essay Scoring Mechanismcited by 7
  8. 2021AES Systems Are Both Overstable And Oversensitive: Explaining Why And Proposing Defensescited by 5
  9. 2021MINIMAL: Mining Models for Data Free Universal Adversarial Triggerscited by 4
  10. 2022Ensembling improves stability and power of feature selection for deep learning modelscited by 3
  11. 2025Ensembling Sparse Autoencoderscited by 3
  12. 2025Robust Explanations Through Uncertainty Decomposition: A Path to Trustworthier AIcited by 2
2 more
  1. 2025Measuring Uncertainty in Transformer Circuits with Effective Information Consistencycited by 1
  2. 2026Aggregate Models, Not Explanations: Improving Feature Importance Estimationcited by 1

Built on by

Cited by year (2026 to 7 Oct)

Citing papers per year: 2026: 1.

1 paper.

  1. 2026Circuit Condensation: Post-Training that Concentrates a Behavior's Causal Circuitcited by 0