AI Paper Atlas

Research map / Vision-language and multimodal / Multimodal LLMs

The First to Know: How Token Distributions Reveal Hidden Knowledge in Large Vision-Language Models?

Qinyu Zhao, Ming Xu, Kartik Gupta, Akshay Asthana, Liang Zheng, Stephen Gould

ECCV 2024 · Vision-language and multimodal › Multimodal LLMs · approach: Bias-aware multi-image robustness · 27 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
train a linear probe on first-token logits instead of using the LVLM's own responseLLaVA-7B (OE prompt) · VizWizAUC: 60.89 -> 90.05Improved
linear probing on first-token logits for jailbreak defenseLLaVA-7B (OE prompt) · MM-SafetyBenchASR: 70.80 -> 7.00Improved
linear probing on first-token logits for deceptive questionsLLaVA-7B (OE prompt) · MAD-BenchASR: 80.00 -> 7.22Improved
use logits of subsequent tokens instead of the first token (ablation)linear probing · VizWiz, MM-SafetyBench, MAD-BenchAUC / ACCHurt
use different prompts (OE, Hint, Meta) with linear probing (ablation)linear probing · three benchmarksAUC / ACCNo clear effect
use hidden states of middle layers instead of logits (ablation)linear probing · three benchmarksAUCNo clear effect
probe CLIP text embeddings alone (ablation)CLIP · MM-SafetyBenchAUC: 95.76 vs LLaVA LP 97.03Improved
increase probe training images from 100 to 1000 (ablation)LLaVA linear probing on POPE adversarial · POPE adversarialACC: 82.77 -> 85.33 (original 83.67)Improved
add LDA on first logits for classificationInstructBLIP · ImageNet-1k valTop-1 accuracy: 55.10 -> 73.71Improved
finetune or retrain LLaVA with the same training data instead of linear probingLLaVA · six taskstask performanceMixed

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