REVIEW 5 cited by
Understanding Multimodal LLMs: the Mechanistic Interpretability of Llava in Visual Question Answering
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
Understanding Multimodal LLMs: the Mechanistic Interpretability of Llava in Visual Question Answering
read the original abstract
Understanding the mechanisms behind Large Language Models (LLMs) is crucial for designing improved models and strategies. While recent studies have yielded valuable insights into the mechanisms of textual LLMs, the mechanisms of Multi-modal Large Language Models (MLLMs) remain underexplored. In this paper, we apply mechanistic interpretability methods to analyze the visual question answering (VQA) mechanisms in the first MLLM, Llava. We compare the mechanisms between VQA and textual QA (TQA) in color answering tasks and find that: a) VQA exhibits a mechanism similar to the in-context learning mechanism observed in TQA; b) the visual features exhibit significant interpretability when projecting the visual embeddings into the embedding space; and c) Llava enhances the existing capabilities of the corresponding textual LLM Vicuna during visual instruction tuning. Based on these findings, we develop an interpretability tool to help users and researchers identify important visual locations for final predictions, aiding in the understanding of visual hallucination. Our method demonstrates faster and more effective results compared to existing interpretability approaches. Code: \url{https://github.com/zepingyu0512/llava-mechanism}
Forward citations
Cited by 5 Pith papers
-
How Do VLMs Fail? Vision-Operation Misalignment in Compositional VQA
The paper proposes four operation-level VLM failure modes and a pathway dissociation, but the dissociation is not supported by the paper's own intervention statistics.
-
Null Space Constrained Contrastive Visual Forgetting for MLLM Unlearning
A contrastive visual forgetting technique constrained to the null space of retained knowledge enables targeted unlearning of visual concepts in MLLMs while preserving non-target visual and all textual knowledge.
-
Locate-then-Sparsify: Attribution Guided Sparse Strategy for Visual Hallucination Mitigation
LTS-FS locates hallucination-relevant layers in LVLMs via causal attribution on a constructed dataset and applies sparse layerwise feature steering to mitigate hallucinations while preserving general task performance.
-
From Heads to Neurons: Causal Attribution and Steering in Multi-Task Vision-Language Models
HONES ranks feed-forward neurons by their causal contributions from task-relevant attention heads and uses lightweight scaling to steer performance on multiple vision-language tasks.
-
Multi-Modal LLM based Image Captioning in ICT: Bridging the Gap Between General and Industry Domain
A 7B-parameter domain-specific image captioning model for ICT, trained in three stages on synthesized and annotated data, outperforms 32B-parameter general models on BLEU and expert accuracy metrics.
discussion (0)
Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.