REVIEW 11 cited by
From Redundancy to Relevance: Information Flow in LVLMs Across Reasoning Tasks
Not yet reviewed by Pith; the record is open.
This paper has not been read by Pith yet. Machine review is queued; the pith claim, tier, and objections will appear here once it completes.
SPECIMEN: schema-true, not a live event
T0 review · schema-true
One-sentence machine reading of the paper's core claim.
pith:XXXXXXXX · record.json · timestamp
Signed reviews
read the original abstract
Large Vision Language Models (LVLMs) achieve great performance on visual-language reasoning tasks, however, the black-box nature of LVLMs hinders in-depth research on the reasoning mechanism. As all images need to be converted into image tokens to fit the input format of large language models (LLMs) along with natural language prompts, sequential visual representation is essential to the performance of LVLMs, and the information flow analysis approach can be an effective tool for determining interactions between these representations. In this paper, we propose integrating attention analysis with LLaVA-CAM, concretely, attention scores highlight relevant regions during forward propagation, while LLaVA-CAM captures gradient changes through backward propagation, revealing key image features. By exploring the information flow from the perspective of visual representation contribution, we observe that it tends to converge in shallow layers but diversify in deeper layers. To validate our analysis, we conduct comprehensive experiments with truncation strategies across various LVLMs for visual question answering and image captioning tasks, and experimental results not only verify our hypothesis but also reveal a consistent pattern of information flow convergence in the corresponding layers, and the information flow cliff layer will be different due to different contexts. The paper's source code can be accessed from \url{https://github.com/zhangbaijin/From-Redundancy-to-Relevance}
Forward citations
Cited by 11 Pith papers
-
Seeing It or Not? Interpretable Vision-aware Latent Steering to Mitigate Object Hallucinations
VaLSe uses attention-based visual contribution maps to steer an LVLM's latent features toward visually grounded content, reducing object hallucinations on several benchmarks while exposing flaws in CHAIR-style evaluation.
-
SurgXBench: Explainable Vision-Language Model Benchmark for Surgery
An explainability-based benchmark showing that surgical vision-language models often make correct predictions without attending to the relevant instruments or tissue.
-
Analyzing Finetuning Representation Shift for Multimodal LLMs Steering
Concept shift vectors, computed as mean activation differences, can partially recover fine-tuned multimodal LLM concepts and steer model outputs without additional training.
-
Enhancing Multimodal Large Language Models Complex Reason via Similarity Computation
Simignore improves multimodal LLM complex question answering on ScienceQA by masking image tokens whose embeddings have low cosine similarity to the text prompt.
-
Cross-modal Information Flow in Multimodal Large Language Models
In LLaVA multimodal models, visual information flows into question token representations in two stages, global then object-specific, before propagating to the final answer position.
-
Toward Modality Gap: Vision Prototype Learning for Weakly-supervised Semantic Segmentation with CLIP
Weakly supervised segmentation with CLIP improves by learning per-class prototypes in vision space instead of relying on text prototypes.
-
Seeing Clearly by Layer Two: Enhancing Attention Heads to Alleviate Hallucination in LVLMs
A training-free intervention that broadcasts the densest vision-sink attention head's map to all heads in shallow layers lowers hallucination rates on CHAIR across several LVLMs.
-
MCA-LLaVA: Manhattan Causal Attention for Reducing Hallucination in Large Vision-Language Models
MCA-LLaVA reindexes image tokens by sums of mirrored 2D coordinates so instruction tokens attend across the whole image, reducing hallucination on POPE, CHAIR, and MME.
-
From Visuals to Vocabulary: Establishing Equivalence Between Image and Text Token Through Autoregressive Pre-training in MLLMs
Adding an L2 loss that pushes the language model's image hidden states back toward the input image embeddings improves LLaVA-style models on several VQA benchmarks, with some benchmarks unaffected or slightly worse.
-
Explainable and Interpretable Multimodal Large Language Models: A Comprehensive Survey
A survey maps the field of MLLM explainability and interpretability into data, model, and training and inference perspectives.
-
First-place Solution for Streetscape Shop Sign Recognition Competition
A team reports winning a street-view shop sign recognition competition with a multi-stage OCR pipeline built from known components, but provides no code, data, or rigorous ablations.
Discussion (0). Continue with ORCID to comment.