Through the LENS: Local Geometric Decomposition of Vision-Language Model Representations

• URL: http://arxiv.org/abs/2608.00561v1
• Date: Sat, 01 Aug 2026 09:54:54 GMT
• Title: Through the LENS: Local Geometric Decomposition of Vision-Language Model Representations
• Abstract summary: LENS (Local Explanation of Neighborhood Subspaces) decomposes VLM activations into local low-rank Gaussian neighborhoods.<n>An automated multimodal labeling pipeline assigns concise semantic descriptions to these neighborhoods.<n>Interpolating activations toward neighborhood centroids causally redirects generation within and across modalities.
• Score: 0.5872014229110214
• License: http://arxiv.org/licenses/nonexclusive-distrib/1.0/
• Abstract: Vision-language models (VLMs) process image patches and text tokens in a shared residual stream, but the local geometry through which the two modalities interact remains poorly understood. Most interpretability methods identify global linear directions, which may miss representations that are globally high-dimensional but locally low-dimensional. We introduce LENS (Local Explanation of Neighborhood Subspaces), a method that decomposes VLM activations into local low-rank Gaussian neighborhoods using a Mixture of Factor Analyzers. Applied to LLaVA-1.5-7B and Qwen3-VL-8B, LENS reveals distinct depth-dependent fusion trajectories consistent with each model's fusion mechanism: LLaVA progressively mixes modalities at later layers, whereas Qwen3-VL mixes them early, partially re-segregates them, and recombines them near the output. An automated multimodal labeling pipeline assigns concise semantic descriptions to these neighborhoods. Interpolating activations toward neighborhood centroids causally redirects generation within and across modalities and outperforms difference-in-means and VL-SAE in most evaluated conditions; in one LLaVA vision-to-vision setting, MFA achieves 5.7 times the VL-SAE score. Human evaluation finds MFA steering competitive with prompting and substantially stronger than the other intervention baselines. Finally, the MFA coefficient space improves Qwen3-VL image-to-rendered-text retrieval at the deepest evaluated layer from 14.9% to 48.6% R@1. Ablations show that the reported fusion trajectories are stable across component counts, local ranks, and modality-purity thresholds. These results support local geometric neighborhoods as useful interpretable and causal units for analyzing cross-modal representations in the evaluated VLMs.

Related papers

• Learning Where Outcomes Change:Credit-Addressable Reasoning for Multimodal Geometry [105.89493240787861]
  We introduce credit-addressable reasoning, in which semantic units exposed during inference also define where learning compares alternatives and assigns credit.<n>We instantiate this principle with Code-CoT, which represents visual relations as line-addressable executable code, and organizes reasoning into typed events.<n>Across nine geometry benchmarks, CE-GRPO achieves an average accuracy of 76.04, outperforming Qwen3-VL-8B and trajectory-level GRPO by $8.09$ and 3.43 points, respectively.
  arXiv  Detail & Related papers  (2026-08-31T08:44:36Z)
• Explainable and Resource-Efficient Spatial Reasoning in Multimodal LLMs for Decision-Critical Applications [7.911287368927542]
  Multimodal Large Language Models (MLLMs) are increasingly deployed in decision-critical pipelines such as robotics, embodied AI, and safety monitoring.<n>MLLMs demonstrate strong reasoning but often struggle with fine-grained spatial understanding and object hallucination.<n>We propose ByDeWay-V2, which integrates explicit spatial context alongside depth cues.
  arXiv  Detail & Related papers  (2026-07-29T17:20:56Z)
• Steering Vision-Language Models with Joint Sparse Autoencoders [2.367986397868328]
  We introduce the Joint Sparse Autoencoder (JSAE) to factorize sequence-pooled vision and language activations into shared, interpretable image/caption-level features.<n>Applying JSAE to vision-language models (VLMs) yields representations that are difficult to use as controllable cross-modal steering directions.
  arXiv  Detail & Related papers  (2026-06-24T10:11:04Z)
• From Drop-off to Recovery: A Mechanistic Analysis of Segmentation in MLLMs [50.185593677108436]
  Multimodal Large Language Models (MLLMs) are increasingly applied to pixel-level vision tasks, yet their intrinsic capacity for spatial understanding remains poorly understood.<n>We investigate segmentation capacity through a layerwise linear probing evaluation across the entire MLLM pipeline: vision encoder, adapter, and LLM.
  arXiv  Detail & Related papers  (2026-03-18T00:22:15Z)
• Rethinking VLMs for Image Forgery Detection and Localization [55.32700985102152]
  We study how to fully leverage vision-language models (VLMs) to assist the image forgery detection and localization (IFDL) task.<n>Building on these findings, we propose a new IFDL pipeline named IFDL-VLM.<n>The experimental results show that we consistently achieve new state-of-the-art performance in detection, localization, and interpretability.
  arXiv  Detail & Related papers  (2026-03-13T12:21:31Z)
• BREATH-VL: Vision-Language-Guided 6-DoF Bronchoscopy Localization via Semantic-Geometric Fusion [7.382475458362566]
  We present BREATH-VL, a hybrid framework that integrates semantic cues from vision-language models with geometric information from registration methods for accurate 6-DoF pose estimation.<n>Building on this, our BREATH-VL outperforms state-of-the-art vision-only localization methods in both accuracy and generalization, reducing translational error by 25.5% compared with the best-performing baseline.
  arXiv  Detail & Related papers  (2026-01-07T09:00:52Z)
• Progressive Localisation in Localist LLMs [0.0]
  This paper demonstrates that progressive localization represents the optimal architecture for creating interpretable large language models (LLMs)<n>We investigate whether interpretability constraints can be aligned with natural semantic structure while being applied strategically across network depth.<n>We show that progressive semantic localization, combining semantic block with steep adaptive locality schedules, achieves near-baseline language modeling performance while providing interpretable attention patterns.
  arXiv  Detail & Related papers  (2025-11-23T09:49:13Z)
• HEDGE: Hallucination Estimation via Dense Geometric Entropy for VQA with Vision-Language Models [4.099133096025821]
  Vision-language models (VLMs) enable open-ended visual question answering but remain prone to hallucinations.<n>We present HEDGE, a unified framework for hallucination detection that combines controlled visual perturbations, semantic clustering, and robust uncertainty metrics.
  arXiv  Detail & Related papers  (2025-11-16T17:16:31Z)
• FineRS: Fine-grained Reasoning and Segmentation of Small Objects with Reinforcement Learning [62.11389260206383]
  textscFineRS is a two-stage MLLM-based reinforcement learning framework for segmenting extremely small objects.<n>We present textscFineRS-4k, a new dataset for evaluating MLLMs on attribute-level reasoning and pixel-level segmentation on subtle, small-scale targets.
  arXiv  Detail & Related papers  (2025-10-24T10:14:17Z)
• Unbiased Region-Language Alignment for Open-Vocabulary Dense Prediction [80.67150791183126]
  Pre-trained vision-language models (VLMs) have demonstrated impressive zero-shot recognition capability, but still underperform in dense prediction tasks.<n>We propose DenseVLM, a framework designed to learn unbiased region-language alignment from powerful pre-trained VLM representations.<n>We show that DenseVLM can directly replace the original VLM in open-vocabulary object detection and image segmentation methods.
  arXiv  Detail & Related papers  (2024-12-09T06:34:23Z)
• Hierarchical Matching and Reasoning for Multi-Query Image Retrieval [113.44470784756308]
  We propose a novel Hierarchical Matching and Reasoning Network (HMRN) for Multi-Query Image Retrieval (MQIR) It disentangles MQIR into three hierarchical semantic representations, which is responsible to capture fine-grained local details, contextual global scopes, and high-level inherent correlations. Our HMRN substantially surpasses the current state-of-the-art methods.
  arXiv  Detail & Related papers  (2023-06-26T07:03:56Z)
• LCPFormer: Towards Effective 3D Point Cloud Analysis via Local Context Propagation in Transformers [60.51925353387151]
  We propose a novel module named Local Context Propagation (LCP) to exploit the message passing between neighboring local regions. We use the overlap points of adjacent local regions as intermediaries, then re-weight the features of these shared points from different local regions before passing them to the next layers. The proposed method is applicable to different tasks and outperforms various transformer-based methods in benchmarks including 3D shape classification and dense prediction tasks.
  arXiv  Detail & Related papers  (2022-10-23T15:43:01Z)

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.