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
67 results for “keyword:"harm"” in ICLR 2027. 18 withdrawn or desk rejected listed apart. clearsort:relevanceratingnewestmost read
[PDF]Disagreement in, Disagreement Out? Uneven Downstream Effects of Harm Classifier Differences▼5.8k2.33
[PDF]VMEval: A Multi-Dimensional and Interpretable Framework for Video-Music Harmony Evaluation▲6.1k2.00
[PDF]Privacy Considerations in Multi-Institutional Data Harmonization Through a Category-Theoretic Lens▲6.1k2.00
[PDF]Learning to Stay Safe: Adaptive Regularization Against Safety Degradation during Fine-Tuning▼1361.33
Withdrawn or desk rejected (18)
[PDF]NASA-INR: Non-linear Adaptive Spectrally-spanning Activation for Implicit Neural RepresentationWithdrawn1.00
[PDF]Characterizing When Language Models Take Harmful Actions to Protect Their Assigned GoalsWithdrawn2.00
[PDF]Safety for Whom? Benchmarking Refusal Behavior and Safety Understanding Across Harm Target in Egocentric Multimodal AssistantsWithdrawn1.33