, LLaVA, InstructBLIP), and twelve unlearning methods.
Open-MMUnlearning framework unifies MLLM unlearning evaluation across five benchmarks and eight models.
The authors systematically compare ten unlearning methods and thirteen metrics, finding gradient descent and MIP-Editor achieve the best overall performance, while BLEU scores highest on metric reliability.
Chinese Tech
Junkai Chen · Yuhao He · Qianshan Wei · Junxiang You · Jingwen Shao · Junkai Lin · +16 more
Institute of Automation, Chinese Academy of Sciences · University of the Chinese Academy of Sciences · ByteDance · The Chinese University of Hong Kong · University of Cambridge
Research Digest··2 min read
The authors introduce Open-MMUnlearning, an open-source framework that integrates target-model preparation, multimodal data processing, unlearning, and evaluation for multimodal large language models (MLLMs).
Why this paper
From Institute of Automation, Chinese Academy of Sciences and 11 others
In one line
Open-MMUnlearning unifies benchmarks, models, and methods for standardized MLLM unlearning evaluation.
What we could check
- ·No code link found
- ·No weights link found
- ·No dataset link found
- ·No compute details found
- ·No stated limitations found
- ·No benchmark numbers found
Observed from the paper text and links we have. Absence here means we did not find it, not that it does not exist.
§