Skip to content

When Output Forgetting Is Not Erasure: A Factorial Replicability Study of Contrastive Subnet Erasure #133

Description

@Mokshithdv25

Original article: Selective Amnesia using Contrastive Subnet Erasure for Class Level Unlearning in Vision Models (CVPR 2026)

PDF URL: https://openaccess.thecvf.com/content/CVPR2026/papers/Pramanik_Selective_Amnesia_using_Contrastive_Subnet_Erasure_for_Class_Level_Unlearning_CVPR_2026_paper.pdf
Metadata URL: https://github.com/Mokshithdv25/cse-calibrated-reproduction-study/blob/main/README.md
Code URL: https://github.com/Mokshithdv25/cse-calibrated-reproduction-study

Scientific domain: Machine learning, computer vision, and computational reproducibility
Programming language: Python
Suggested editor:

This submission reports an independent, fail-closed replication and diagnostic study of CSE. It separates targeted channel attenuation from retain-only classifier-head recalibration using factorial controls, feature probes, and cross-dataset validation. The study includes validated positive and negative replication evidence and documents the limits of the original claim.

Activity

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Metadata

Metadata

Assignees

No one assigned

    Type

    No type

    Projects

    No projects

      Milestone

      No milestone

      Relationships

      None yet

      Development

      No branches or pull requests

      Issue actions