AI Paper Atlas

Research map / Label-efficient and robust learning / Knowledge distillation

Token-level Response-visual Attention Guidance for Multimodal LLMs Knowledge Distillation

Jaehyun Jang, Eunseop Yoon, Hee Suk Yoon, SooHwan Eom, Mark Hasegawa-Johnson, Chang D. Yoo

ECCV 2026 · Label-efficient and robust learning › Knowledge distillation · approach: Rethinking distillation objectives · 1 citations (OpenAlex)

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 changedWhereMeasuredEffect
supervise response-to-vision instead of prompt-to-vision attention (ablation)TRAG · Qwen2.5-0.5BVQA/CR Avg: VQA 58.6 -> 60.2; CR 69.2 -> 75.2Improved
add prompt tokens to response supervision (ablation)TRAG · Qwen2.5-0.5BVQA Avg: 58.4 vs 58.6 (prompt only)No clear effect
token-level response supervision instead of aggregated (ablation)TRAG · Qwen2.5-0.5BVQA/CR Avg: VQA 57.7 -> 60.2; CR 70.4 -> 75.2Improved
entropy-weighted forward/reverse KL instead of fixed measures (ablation)TRAG · Qwen2.5-0.5BVQA/CR Avg: VQA 58.7 -> 60.2; CR 74.4 -> 75.2Improved
use intermediate layers [0.3,0.6] instead of early layers (ablation)TRAG · Qwen2.5-0.5BVQA/CR Avg: VQA 58.7 -> 60.2; CR 69.2 -> 75.2Improved
shift aggregated layer range to [0.4,0.7] (ablation)TRAG · Qwen2.5-0.5BVQA/CR Avg: VQA 60.2 -> 60.3; CR 75.2 -> 75.0No clear effect
use larger teacher (4B instead of 1.8B) (ablation)TRAG · Qwen1.5-0.5B studentCR Avg: 63.0 -> 70.3Improved
apply TRAG over PT-SFT on Qwen2.5-0.5B (ablation)TRAG · VQA Avg6Avg6: 62.4 -> 65.3Improved
TRAG versus LLaVA-KD on Qwen2.5-0.5BTRAG · VQAAvg8: 58.7 -> 60.2Improved
TRAG on MobileLLaMA 1.4B studentTRAG · MobileLLaMAVQA/CR Avg: VQA 55.2 -> 58.2; CR 50.0 -> 57.3Improved
TRAG on Qwen2.5-1.5B student on SugarCrepeTRAG · SugarCrepeaccuracy: 87.9 -> 87.3 SugarCrepe; avg 83.8 -> 85.0Mixed

Closest papers by meaning

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.

Open scime draft

Get the next scime Dispatch and a monthly field update

Free. We send a confirmation link first; every email has a one-click unsubscribe. Privacy