UniForget enables joint concept unlearning in multimodal models
Original title: Is Forgetting Unified? Exploring Concept Unlearning in Unified Multimodal Models
UniForget achieves 80.0% joint forgetting on BAGEL-7B-MoT and 92.7% on Janus-Pro-7B, outperforming single-pathway baselines while preserving other capabilities.
Existing unlearning methods fail to suppress target concepts across both generation and understanding pathways in unified multimodal models. UUBench covers 30 target and 60 neighboring concepts; single-pathway approaches yield limited joint forgetting. UniForget is proposed to deliver strong joint forgetting rates without major capability loss.
- 80.0% joint forgetting on BAGEL-7B-MoT
- 92.7% joint forgetting on Janus-Pro-7B
- 30 target concepts in UUBench
- Single-pathway methods limited to one pathway
- Trade-offs between forgetting and preservation
From the original article
1 min read at huggingface.co
Unified multimodal models (UMMs) integrate image generation and visual understanding, but are existing single-pathway unlearning methods sufficient to suppress target concepts across both? In this work, we study cross-pathway concept unlearning. Our goal is to prevent a model from both generating images of a target concept and identifying it in input images, while preserving other capabilities.
By Zhongkai Xue, Junkai Chen, Mingcong Lei · huggingface.co
A short excerpt; the article and its images belong to the publisher. Summary, key facts and assessment above are AI-written from the article and can be wrong.