New unlearning method claims 77x speedup for targeted behavior removal
Influence-weighted unlearning could sharply cut the cost of correcting model behavior if it holds up, but the numbers are the authors' own and await third-party reproduction.
Original event 2026-09-24
Version 4 of the RapidUn preprint, updated September 24, proposes influence-weighted LoRA unlearning. The authors report that on Llama-3-8B, removing targeted trigger behaviors runs 77x faster in wall-clock time than a clean-corpus LoRA retraining reference.
Machine unlearning means removing a specific learned behavior without retraining the whole model. The method converts cross-sample influence estimates into fixed sample weights; the authors report lower trigger attack success rates than the existing Fisher, GA, and LoReUn methods with clean utility largely intact, plus cross-model validation on Mistral-7B.
The baselines are comparable unlearning methods, not full retraining, and the benchmarks were built and run by the authors themselves. Code is released, but no third-party reproduction exists yet.