NeurIPS 2026 reviews are up: reviews, scores, rebuttals and meta-reviews for 7,877 papers. See the highest rated
acceptodds: the papers of ICLR 2027
22 results for “keyword:"activation interventions"” in ICLR 2027. 4 withdrawn or desk rejected listed apart. clearsort:relevanceratingnewestmost read
[PDF]Evidence States Survive Where Answers Fail: Decodability and Causal Access in Language Models▼4.3k2.00
[PDF]From Hallucinated Targets to Runaway Repetition: Termination Control in Long Structured Generation▼5.9k2.50
[PDF]INTERVENTION IS NOT IDENTIFICATION: ACTIVATION EDITS CAN PRODUCE THE PREDICTED EFFECT FOR THE WRONG REASON▼1351.33
[PDF]Overtopping in LLMs: Spiking-Like Causal Control and the Limits of Static Intervention▼5.8k2.33
[PDF]Content-Dependent Recovery from Task-Relevant Activation Disruption in Large Language Models▲4012.67
[PDF]Information Anchors: Understanding and Using Forecast-Relevant Information in Time-Series Foundation Models▼5.8k2.33
Withdrawn or desk rejected (4)
[PDF]Calibrating Causal-Localization Scores: Feature-space comparisons depend on intervention budget and write-back semanticsWithdrawn2.00