Optimizing Multimodal Language Models through Attention-based Interpretability

• URL: http://arxiv.org/abs/2511.23375v1
• Date: Fri, 28 Nov 2025 17:21:31 GMT
• Title: Optimizing Multimodal Language Models through Attention-based Interpretability
• Authors: Alexander Sergeev, Evgeny Kotelnikov,
• Abstract summary: Fine-tuning multimodal language models is computationally expensive.<n>We propose an attention-head interpretability method by analyzing attention scores relative to image key objects.<n>We conduct experiments on tokens with 2-3 billion parameters to validate the method's effectiveness.
• Score: 45.88028371034407
• License: http://creativecommons.org/licenses/by-nc-sa/4.0/
• Abstract: Modern large language models become multimodal, analyzing various data formats like text and images. While fine-tuning is effective for adapting these multimodal language models (MLMs) to downstream tasks, full fine-tuning is computationally expensive. Parameter-Efficient Fine-Tuning (PEFT) methods address this by training only a small portion of model weights. However, MLMs are difficult to interpret, making it challenging to identify which components are most effective for training to balance efficiency and performance. We propose an attention-based interpretability method for MLMs by analyzing attention scores relative to image tokens. The core idea is to identify attention heads that focus on image key objects. We utilize this information to select optimal model components for PEFT in multimodal models. Our contributions include a method for identifying attention heads associated with image key objects, its application to PEFT for image captioning, and the creation of a new dataset containing images, key object masks, and their textual descriptions. We conducted experiments on MLMs with 2-3 billion parameters to validate the method's effectiveness. By calculating Head Impact (HI) scores we quantify an attention head's focus on key objects, indicating its significance in image understanding. Our fine-tuning experiments demonstrate that adapting layers with the highest HI scores leads to the most significant shifts in metrics compared to pre-trained, randomly selected, or lowest-HI-score layers. This indicates that fine-tuning a small percentage (around 0.01%) of parameters in these crucial layers can substantially influence image understanding capabilities.

Related papers

• When Visualizing is the First Step to Reasoning: MIRA, a Benchmark for Visual Chain-of-Thought [118.71264263478083]
  We propose MIRA, a new benchmark designed to evaluate models in scenarios where generating intermediate visual images is essential for successful reasoning.<n>We include 546 multimodal problems, annotated with intermediate visual images and final answers.
  arXiv  Detail & Related papers  (2025-11-04T18:00:51Z)
• Towards Text-Image Interleaved Retrieval [49.96332254241075]
  We introduce the text-image interleaved retrieval (TIIR) task, where the query and document are interleaved text-image sequences.<n>We construct a TIIR benchmark based on naturally interleaved wikiHow tutorials, where a specific pipeline is designed to generate interleaved queries.<n>We propose a novel Matryoshka Multimodal Embedder (MME), which compresses the number of visual tokens at different granularity.
  arXiv  Detail & Related papers  (2025-02-18T12:00:47Z)
• PiLaMIM: Toward Richer Visual Representations by Integrating Pixel and Latent Masked Image Modeling [7.630967411418269]
  We propose PiLaMIM, a unified framework that combines Pixel MIM and Latent MIM to integrate their complementary strengths.<n>Our method uses a single encoder along with two distinct decoders: one for predicting pixel values and another for latent representations, ensuring the capture of both high-level and low-level visual features.
  arXiv  Detail & Related papers  (2025-01-06T13:30:16Z)
• Modality-Fair Preference Optimization for Trustworthy MLLM Alignment [22.093944381988496]
  Multimodal large language models (MLLMs) have achieved remarkable success across various tasks.<n>However, separate training of visual and textual encoders often results in a misalignment of the modality.<n>These inaccuracies severely undermine the trustworthiness of MLLMs in real-world applications.
  arXiv  Detail & Related papers  (2024-10-20T08:56:52Z)
• Enhancing Large Vision Language Models with Self-Training on Image Comprehension [131.14381425260706]
  We introduce Self-Training on Image (STIC), which emphasizes a self-training approach specifically for image comprehension. First, the model self-constructs a preference for image descriptions using unlabeled images. To further self-improve reasoning on the extracted visual information, we let the model reuse a small portion of existing instruction-tuning data.
  arXiv  Detail & Related papers  (2024-05-30T05:53:49Z)
• PTA: Enhancing Multimodal Sentiment Analysis through Pipelined Prediction and Translation-based Alignment [17.70859235594373]
  Multimodal aspect-based sentiment analysis (MABSA) aims to understand opinions in a granular manner. Traditionally, MABSA methods use a joint prediction approach to identify aspects and sentiments simultaneously. We propose a pipeline framework that first predicts the aspect and then uses translation-based alignment (TBA) to enhance multimodal semantic consistency for better image utilization.
  arXiv  Detail & Related papers  (2024-05-23T01:16:45Z)
• MM1: Methods, Analysis & Insights from Multimodal LLM Pre-training [103.72844619581811]
  We build performant Multimodal Large Language Models (MLLMs) In particular, we study the importance of various architecture components and data choices. We demonstrate that for large-scale multimodal pre-training using a careful mix of image-caption, interleaved image-text, and text-only data.
  arXiv  Detail & Related papers  (2024-03-14T17:51:32Z)
• Browse and Concentrate: Comprehending Multimodal Content via prior-LLM Context Fusion [70.9767518332692]
  Multimodal Large Language Models (MLLMs) that incorporate LLMs with pre-trained vision models have recently demonstrated impressive performance across diverse vision-language tasks. However, they fall short to comprehend context involving multiple images. We propose a two phase paradigm, browse-and-concentrate, to enable in-depth multimodal context fusion.
  arXiv  Detail & Related papers  (2024-02-19T14:59:07Z)
• MOCA: Self-supervised Representation Learning by Predicting Masked Online Codebook Assignments [72.6405488990753]
  Self-supervised learning can be used for mitigating the greedy needs of Vision Transformer networks. We propose a single-stage and standalone method, MOCA, which unifies both desired properties. We achieve new state-of-the-art results on low-shot settings and strong experimental results in various evaluation protocols.
  arXiv  Detail & Related papers  (2023-07-18T15:46:20Z)

This list is automatically generated from the titles and abstracts of the papers in this site.

This site does not guarantee the quality of this site (including all information) and is not responsible for any consequences.