RoCo-ACE: Rollout-Conditioned Online Distillation for Retention-Aware Knowledge Injection
Quick Answer
RoCo-ACE introduces a novel rollout-conditioned online distillation method for knowledge injection in MLLMs, achieving superior accuracy in knowledge retention while minimizing behavioral drift.
Quick Take
It reallocates distillation weight to reference-supported tokens and employs sparse corrections for omitted authoritative anchors, outperforming existing methods across multiple benchmarks.
Key Points
- RoCo-ACE improves knowledge injection accuracy while maintaining retention close to base models.
- Utilizes reference-free and reference-conditioned likelihood contrast for effective distillation.
- Introduces sparse corrections for authoritative information not captured in rollouts.
- Demonstrated effectiveness across three knowledge-injection settings and six retention benchmarks.
- Outperforms multiple baselines and models in injected-knowledge accuracy.
Paper Resources
Source Excerpt
Knowledge injection updates pretrained MLLMs with new factual or domain-specific knowledge, but fitting full authoritative answers can cause drift in non-updated behavior. Online distillation mitigates this drift by training on model-generated rollouts, yet uniform reference-conditioned distillation provides coarse supervision: it can under-emphasize reference-supported rollout tokens and supervise omitted facts only indirectly. We introduce RoCo-ACE, a rollout-conditioned online distillation ob
Want this in your inbox every morning?
Daily brief at your local 8am — bilingual EN/中文, free.
More from arXiv cs.AI
See more →HOBA: Hierarchical On-Policy Bidding Agents for Adaptive Online Advertising
HOBA (Hierarchical On-policy Bidding Agents) is a novel hierarchical reinforcement learning framework that enhances online advertising bidding systems by improving adaptability and reducing hyperparameter tuning costs. It utilizes a for hyperparameter inference, a SARSA agent for expert model selection, and a dynamic expert pool for bid execution, achieving a +3.6% increase in target cost during large-scale deployment and outperforming state-of-the-art baselines on AuctionNet.