Your CLIP has 164 dimensions of noise: Exploring the embeddings covariance eigenspectrum of contrastively pretrained vision-language transformers

• URL: http://arxiv.org/abs/2605.14893v1
• Date: Thu, 14 May 2026 14:37:50 GMT
• Title: Your CLIP has 164 dimensions of noise: Exploring the embeddings covariance eigenspectrum of contrastively pretrained vision-language transformers
• Abstract summary: Pre-trained Vision-Language Models (VLMs) serve as powerful feature extractors.<n>Their shared latent spaces are prone to structural anomalies and act as repositories for non-semantic, multi-modal noise.<n>We decompose the VLM latent space into a multi-modal semantic signal component and a shared noise subspace.
• Score: 0.764671395172401
• License: http://creativecommons.org/licenses/by/4.0/
• Abstract: Contrastively pre-trained Vision-Language Models (VLMs) serve as powerful feature extractors. Yet, their shared latent spaces are prone to structural anomalies and act as repositories for non-semantic, multi-modal noise. To address this phenomenon, we employ spectral decomposition of covariance matrices to decompose the VLM latent space into a multi-modal semantic signal component and a shared noise subspace. We observe that this noise geometry exhibits strong subgroup invariance across distinct data subsets. Crucially, pruning these shared noise dimensions is mainly harmless, preserving or actively improving downstream task performance. By isolating true semantic signals from artifactual noise, this work provides new mechanistic insights into the representational structure of modern VLMs, suggesting that a substantial fraction of their latent geometry is governed by shared, architecture-level noise rather than task-relevant semantics alone.

Related papers

• Dynamic-Robust Photometric-Semantic Reconstruction for Open-Vocabulary 3D Scene Understanding [63.599342750566485]
  SPAR is a novel joint semantic-geometric encoding architecture that explicitly isolates transient dynamic noise prior to latent space aggregation.<n>Our end-to-end approach achieves exceptional novel view synthesis quality, yielding a PSNR of 22.15 dB and 23.33 dB given only 3 and 4 input views respectively.
  arXiv  Detail & Related papers  (2026-08-29T09:58:24Z)
• MIND: Decoupling Model-Induced Label Noise via Latent Manifold Disentanglement [3.626013617212667]
  Model-induced label noise dominates noise in classical robust learning.<n>We propose Model-Induced Noise Decoupling (MIND), a theoretically grounded framework addressing this dilemma.<n>MIND significantly outperforms state-of-the-art methods on complex benchmarks.
  arXiv  Detail & Related papers  (2026-05-15T15:44:19Z)
• Multimodal Structure Learning: Disentangling Shared and Specific Topology via Cross-Modal Graphical Lasso [12.238577096609761]
  We propose Cross-Modal Graphical Lasso (CM-GLasso) for learning interpretable multimodal representations.<n>By coupling a novel text-visualization strategy with a unified vision-language encoder, we strictly align multimodal features into a shared latent space.<n>We show that CM-GLasso establishes a new state-of-the-art in generative classification and dense semantic segmentation tasks.
  arXiv  Detail & Related papers  (2026-04-05T04:07:59Z)
• Spectral Disentanglement and Enhancement: A Dual-domain Contrastive Framework for Representation Learning [28.392130815615545]
  Spectral Disentanglement and Enhancement (SDE) is a novel framework that bridges the gap between the geometry of the embedded spaces and their spectral properties.<n>SDE consistently improves representation and robustness, outperforming state-of-the-art methods.
  arXiv  Detail & Related papers  (2026-02-09T07:29:43Z)
• Explaining multimodal LLMs via intra-modal token interactions [55.27436637894534]
  Multimodal Large Language Models (MLLMs) have achieved remarkable success across diverse vision-language tasks, yet their internal decision-making mechanisms remain insufficiently understood.<n>We propose enhancing interpretability by leveraging intra-modal interaction.
  arXiv  Detail & Related papers  (2025-09-26T14:39:13Z)
• Implicit Counterfactual Learning for Audio-Visual Segmentation [50.69377287012591]
  We propose the implicit counterfactual framework (ICF) to achieve unbiased cross-modal understanding.<n>Due to the lack of semantics, heterogeneous representations may lead to erroneous matches.<n>We introduce the multi-granularity implicit text (MIT) involving video-, segment- and frame-level as the bridge to establish the modality-shared space.
  arXiv  Detail & Related papers  (2025-07-28T11:46:35Z)
• Be Decisive: Noise-Induced Layouts for Multi-Subject Generation [56.80513553424086]
  Complex prompts lead to subject leakage, causing inaccuracies in quantities, attributes, and visual features.<n>We introduce a new approach that predicts a spatial layout aligned with the prompt, derived from the initial noise, and refines it throughout the denoising process.<n>Our method employs a small neural network to predict and refine the evolving noise-induced layout at each denoising step.
  arXiv  Detail & Related papers  (2025-05-27T17:54:24Z)
• FreSca: Scaling in Frequency Space Enhances Diffusion Models [55.75504192166779]
  This paper explores frequency-based control within latent diffusion models.<n>We introduce FreSca, a novel framework that decomposes noise difference into low- and high-frequency components.<n>FreSca operates without any model retraining or architectural change, offering model- and task-agnostic control.
  arXiv  Detail & Related papers  (2025-04-02T22:03:11Z)
• QDFormer: Towards Robust Audiovisual Segmentation in Complex Environments with Quantization-based Semantic Decomposition [47.103732403296654]
  Multi-source semantic space can be represented as the Cartesian product of single-source sub-spaces. We introduce a global-to-local quantization mechanism, which distills knowledge from stable global (clip-level) features into local (frame-level) ones. Experiments demonstrate that our semantically decomposed audio representation significantly improves AVS performance.
  arXiv  Detail & Related papers  (2023-09-29T20:48:44Z)
• Robust Semantic Communications with Masked VQ-VAE Enabled Codebook [56.63571713657059]
  We propose a framework for the robust end-to-end semantic communication systems to combat the semantic noise. To combat the semantic noise, the adversarial training with weight is developed to incorporate the samples with semantic noise in the training dataset. We develop a feature importance module (FIM) to suppress the noise-related and task-unrelated features.
  arXiv  Detail & Related papers  (2022-06-08T16:58:47Z)
• LAAT: Locally Aligned Ant Technique for discovering multiple faint low dimensional structures of varying density [0.0]
  In manifold learning, several studies indicate solutions for removing background noise or noise close to the structure when the density is substantially higher than that exhibited by the noise. We propose a novel method to extract manifold points in the presence of noise based on the idea of Ant colony optimization. In contrast to the existing random walk solutions, our technique captures points that are locally aligned with major directions of the manifold.
  arXiv  Detail & Related papers  (2020-09-17T14:22:50Z)

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.