Unmerge: Efficient Machine Unlearning via Task Arithmetic

arXiv:2609.38895 · cs.LG, cs.AI · Submitted 2026-09-30 · Read on arXiv

cs.LG, cs.AI

Submitted: 2026-09-30

Updated: 2026-09-30

Code: https://github.com/naver-ai/negmerge

Terminology

• Machine Unlearning — 1× in this paper · explained in 3 episodes

Sources

• OpenUnlearning: Accelerating LLM Unlearning via Unified Benchmarking of Methods and Metrics
• The Llama 3 Herd of Models
• Unlearning's Blind Spots: Over-Unlearning and Prototypical Relearning Attack
• Class Unlearning via Depth-Aware Removal of Forget-Specific Directions
• NegMerge: Sign-Consensual Weight Merging for Machine Unlearning
• Exact Unlearning of Finetuning Data via Model Merging at Scale
• TOFU: A Task of Fictitious Unlearning for LLMs
• Is your algorithm unlearning or untraining?
• Machine Unlearning: A Comprehensive Survey

Related papers

• Polynomial-Augmented Neural Networks (PANNs) with Weak Orthogonality Constraints for Enhanced Function and PDE Approximation
• AIRL-S: Unifying Reinforcement Learning and Search-Based Test-Time Scaling via Adversarial Inverse Reinforcement Learning
• Transformers as Bayesian In-Context Experimenters: Smoothness-Adaptive Efficient ATE Estimation
• Convergence issues in Relational Concept Analysis based on AOC-posets
• Beliefs Beyond Posteriors: The Predictive Cost of ELBO Binding
• Understanding Diffusion Models via Ratio-Based Function Approximation with SignReLU Networks