Challenge: Existing work on causal interpretability focuses on large language models (LLMs) but internal mechanisms of vision-language models remain underexplored, authors say .
Approach: They introduce a framework that combines visual and semantic manipulations for causal interpretation of vision-language models.
Outcome: The proposed framework shows improved performance for LLAVA and InstructBLIP on three diverse benchmarks.

Similar Papers

Treble Counterfactual VLMs: A Causal Approach to Hallucination (2025.findings-emnlp)

Copied to clipboard

Challenge: Existing studies link hallucination to data or representation biases, but their causal origins remain unclear.
Approach: They propose a causal framework to analyze and mitigate hallucination in vision-language models by using counterfactual analysis to estimate the Natural Direct Effect (NDE) of each modality and their interaction.
Outcome: The proposed framework significantly reduces hallucination while preserving task performance while retaining reliability.
What Do VLMs NOTICE? A Mechanistic Interpretability Pipeline for Gaussian-Noise-free Text-Image Corruption and Evaluation (2025.naacl-long)

Copied to clipboard

Challenge: Vision-Language Models (VLMs) have gained prominence due to their success in solving complex cross-modal tasks.
Approach: They propose a Gaussian-Noise-free pipeline for mechanistic interpretability in VLMs that introduces Semantic Image Pairs corruption, the first visual counterpart to Symmetric Token Replacement for text.
Outcome: The proposed pipeline identifies a set of “universal attention heads” in BLIP and LLaVA that consistently contribute across different tasks and modalities.
What’s Missing in Vision-Language Models? Probing Their Struggles with Causal Order Reasoning (2026.eacl-long)

Copied to clipboard

Challenge: Existing benchmarks often include a mixture of reasoning questions, making it difficult to truly assess VLMs’ causal reasoning abilities.
Approach: They propose two new benchmarks specifically designed to isolate and rigorously evaluate VLMs’ causal reasoning abilities.
Outcome: The proposed benchmarks show that vision-language models perform poorly on causal reasoning tasks, often only marginally surpassing random guessing.
Weaving Context Across Images: Improving Vision-Language Models through Focus-Centric Visual Chains (2025.acl-long)

Copied to clipboard

Challenge: Existing vision-language models struggle to disentangle information scattered across complex visual inputs, leading to performance degradation.
Approach: They propose a focus-centric visual chain paradigm that enhances VLMs’ perception, comprehension, and reasoning abilities in multi-image scenarios.
Outcome: The proposed approach achieves average performance gains of 3.16% and 2.24% across two distinct model architectures, without compromising the general vision-language capabilities.
Can VLMs Actually See and Read? A Survey on Modality Collapse in Vision-Language Models (2025.findings-acl)

Copied to clipboard

Challenge: Vision-language models integrate textual and visual information, enabling them to process visual inputs and generate predictions.
Approach: They review work on modality collapse analysis to provide insights into the reason for this unintended behavior and review probing studies for fine-grained vision-language understanding.
Outcome: The proposed models can achieve competitive performance in vision-language tasks despite relying heavily on textual information and ignoring visual information.
CausalVLBench: Benchmarking Visual Causal Reasoning in Large Vision-Language Models (2025.emnlp-main)

Copied to clipboard

Challenge: Large vision-language models have shown impressive ability in various language tasks, especially with their emergent in-context learning capability.
Approach: They propose a causal reasoning benchmark for multi-modal in-context learning from large vision-language models that incorporates visual inputs.
Outcome: The proposed model outperforms existing models on three visual causal reasoning tasks and demonstrates their strengths and weaknesses.
Adapting Vision-Language Models for E-commerce Understanding at Scale (2026.eacl-industry)

Copied to clipboard

Challenge: Existing approaches to adapt VLMs to attribute-centric, multi-image, and noisy data are limited.
Approach: They propose a novel evaluation suite that incorporates deep product understanding, strict instruction following, and dynamic attribute extraction.
Outcome: The proposed model improves e-commerce performance while preserving broad multimodal capabilities.
A Corpus for Visual Question Answering Annotated with Frame Semantic Information (2020.lrec-1)

Copied to clipboard

Challenge: Visual Question Answering (VQA) is a computer vision problem.
Approach: They propose to annotate a visual question answering dataset with verb semantics to help the model understand verbs.
Outcome: The proposed system is built on the imSitu dataset annotated with verb semantic information.
How Do LLMs and VLMs Understand Viewpoint Rotation Without Vision? An Interpretability Study (2026.acl-long)

Copied to clipboard

Challenge: Existing studies on spatial intelligence from the perspective of visual-spatial intelligence have not explored whether visual intelligence alone is sufficient to endow models with spatial intelligence.
Approach: They propose to use a linguistic perspective to investigate spatial intelligence from a theoretical perspective.
Outcome: The proposed model performs poorly on the proposed dataset while human can easily achieve 100% accuracy.
Guiding Medical Vision-Language Models with Diverse Visual Prompts: Framework Design and Comprehensive Exploration of Prompt Variations (2025.naacl-long)

Copied to clipboard

Challenge: Current vision-language models lack the ability to focus on specific areas designated by humans . a new framework that integrates medical entity extraction, visual prompt generation, and dataset adaptation is proposed to improve visual prompt-guided fine-tuning.
Approach: They propose to use visual prompts to guide and enhance formation of region-specific attention.
Outcome: The proposed framework outperforms state-of-the-art large vision-language models on medical datasets.

What is GenGO?

GenGO is an NLP powered publication search system. It currenctly indexes 30k+ papers from ACL Anthology, and implements multi-aspect summarization, semantic search, and more!

Information

About
Limitations