UUBench: Cross-Pathway Concept Unlearning in Unified Multimodal Models
CASIA · hf · 2026-11-04
CASIA, ByteDance and collaborators study cross-pathway concept unlearning in unified multimodal models: suppressing both generation and recognition of a target concept while preserving other capabilities. They introduce UUBench, a benchmark covering 30 target and 60 semantically neighboring concepts, show existing single-pathway methods achieve limited joint forgetting, and propose a method that jointly suppresses both pathways.
More from Research
- Harvard Professors Admit AI Article Reviews Now Beat the Best Humans — iamfakhrealam · 2026-11-04
- Pre-pretraining gains traced to embedding init: N(0,0.02) alone lifts perplexity 9% — LauraRuis · 2026-11-04
- kalomaze shares loss curve of larger-backbone run on DeepSeek V3.1 base, eval loss down to ~0.82 — kalomaze · 2026-11-04
- Hobbyist rewrites Qwen 0.6B weights with 39.8M tokens, no gradients, in 3h12m — Altruistic_Heat_9531 · 2026-11-04
- SpecPilot agent boosts speculative decoding from 4.46× to 10.87×, no retraining — pmttyji · 2026-11-04
- Autonomous agent designs target-specific MSAs to improve AlphaFold 3 complex prediction — SJTU · 2026-11-04