Research map / Continual learning / Continual fine-tuning of LLMs
KORE: Enhancing Knowledge Injection for Large Multimodal Models via Knowledge-Oriented Controls
Kailin Jiang, Hongbo Jiang, Ning Jiang, Zhi Gao, Jinhe Bi, Yuchen Ren, Bin Li, Yuntao Du, Lei Liu, Qing Li
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| add KORE-Augmentation (multi-round dialogue and instruction data) (ablation) | KORE · EVOKE (LLaVA-v1.5 7B) | K.A: 14.57 -> 35.96 | Improved |
| add KORE-Constraint (null-space adapter initialization) (ablation) | KORE · EVOKE (LLaVA-v1.5 7B) | K.R: 35.78 -> 40.00 | Improved |
| add KORE-Constraint (null-space adapter initialization) (ablation) | KORE · EVOKE (LLaVA-v1.5 7B) | K.A: 38.82 -> 35.96 | No clear effect |
| freeze matrix A in the null space (ablation) | KORE · EVOKE (LLaVA-v1.5 7B) | K.R: 38.92 -> 40.00 | Improved |
| use KORE-Augmentation instead of knowledge-aware text augmentation (ablation) | KORE · EVOKE (LLaVA-v1.5 7B) | K.A: 20.29 -> 38.82 | Improved |
| use KORE-Augmentation instead of knowledge-agnostic text augmentation (ablation) | KORE · EVOKE (LLaVA-v1.5 7B) | K.A: 15.60 -> 38.82 | Improved |
| use KORE-Augmentation instead of image augmentation (knowledge-aware) (ablation) | KORE · EVOKE (LLaVA-v1.5 7B) | HARS: 66.48 -> 79.04 | Improved |
| increase LoRA rank (ablation) | KORE · EVOKE (LLaVA-v1.5 7B) | overall performance | Improved |
| increase rank from 235 to 256 (ablation) | KORE · EVOKE (LLaVA-v1.5 7B) | Avg: 37.98 -> 39.14 | Improved |
| use KORE instead of Replay on larger model | LLaVA-v1.5 (13B) · EVOKE + retention | Avg: 30.91 -> 42.01 | Improved |
| use KORE instead of Replay on Qwen2.5-VL (7B) | Qwen2.5-VL (7B) · EVOKE + retention | Avg: 39.53 -> 42.72 | Improved |
| use KORE instead of Replay in retention | LLaVA-v1.5 (7B) · EVOKE (LLaVA-v1.5 7B) | K.R: 43.00 -> 40.00 | Hurt |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 1 of 1 on most shared results.
- SEFE — SEFE: Superficial and Essential Forgetting Eliminator for Multimodal Continual Instruction Tuning (ICML 2025) · better on 4 of 4 shared results
Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.
Closest papers by meaning
- Towards Unified Multimodal Editing with Enhanced Knowledge Collaboration (NeurIPS 2024 · similarity 0.88)
- LoRASculpt: Sculpting LoRA for Harmonizing General and Specialized Knowledge in Multimodal Large Language Models (CVPR 2025 · similarity 0.87)
- Cross-Modal Unlearning via Influential Neuron Path Editing in Multimodal Large Language Models (AAAI 2026 · similarity 0.87)
- ComprehendEdit: A Comprehensive Dataset and Evaluation Framework for Multimodal Knowledge Editing (AAAI 2025 · similarity 0.85)
- Active Data Curation Effectively Distills Large-Scale Multimodal Models (CVPR 2025 · similarity 0.85)
- MemEIC: A Step Toward Continual and Compositional Knowledge Editing (NeurIPS 2025 · similarity 0.85)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.
