Interpreting CLIP with Hierarchical Sparse Autoencoders

• URL: http://arxiv.org/abs/2502.20578v1
• Date: Thu, 27 Feb 2025 22:39:13 GMT
• Title: Interpreting CLIP with Hierarchical Sparse Autoencoders
• Authors: Vladimir Zaigrajew, Hubert Baniecki, Przemyslaw Biecek,
• Abstract summary: Matryoshka SAE (MSAE) learns hierarchical representations at multiple granularities simultaneously.<n>MSAE establishes a new state-of-the-art frontier between reconstruction quality and sparsity for CLIP.
• Score: 8.692675181549117
• License: http://arxiv.org/licenses/nonexclusive-distrib/1.0/
• Abstract: Sparse autoencoders (SAEs) are useful for detecting and steering interpretable features in neural networks, with particular potential for understanding complex multimodal representations. Given their ability to uncover interpretable features, SAEs are particularly valuable for analyzing large-scale vision-language models (e.g., CLIP and SigLIP), which are fundamental building blocks in modern systems yet remain challenging to interpret and control. However, current SAE methods are limited by optimizing both reconstruction quality and sparsity simultaneously, as they rely on either activation suppression or rigid sparsity constraints. To this end, we introduce Matryoshka SAE (MSAE), a new architecture that learns hierarchical representations at multiple granularities simultaneously, enabling a direct optimization of both metrics without compromise. MSAE establishes a new state-of-the-art Pareto frontier between reconstruction quality and sparsity for CLIP, achieving 0.99 cosine similarity and less than 0.1 fraction of variance unexplained while maintaining ~80% sparsity. Finally, we demonstrate the utility of MSAE as a tool for interpreting and controlling CLIP by extracting over 120 semantic concepts from its representation to perform concept-based similarity search and bias analysis in downstream tasks like CelebA.

Related papers

• Semi-supervised Semantic Segmentation with Multi-Constraint Consistency Learning [81.02648336552421]
  We propose a Multi-Constraint Consistency Learning approach to facilitate the staged enhancement of the encoder and decoder. Self-adaptive feature masking and noise injection are designed in an instance-specific manner to perturb the features for robust learning of the decoder. Experimental results on Pascal VOC2012 and Cityscapes datasets demonstrate that our proposed MCCL achieves new state-of-the-art performance.
  arXiv  Detail & Related papers  (2025-03-23T03:21:33Z)
• Semi-KAN: KAN Provides an Effective Representation for Semi-Supervised Learning in Medical Image Segmentation [2.717521115234258]
  Semi-supervised medical image segmentation (SSMIS) offers a viable alternative to CNNs and ViTs. Inspired by Kolmogorov-Arnold Networks (KANs), we propose Semi-KAN. KANs exhibit superior representation learning capabilities with fewer parameters. We show that Semi-KAN surpasses baseline networks, utilizing fewer KAN layers and lower computational cost.
  arXiv  Detail & Related papers  (2025-03-19T08:27:41Z)
• "Principal Components" Enable A New Language of Images [79.45806370905775]
  We introduce a novel visual tokenization framework that embeds a provable PCA-like structure into the latent token space. Our approach achieves state-of-the-art reconstruction performance and enables better interpretability to align with the human vision system.
  arXiv  Detail & Related papers  (2025-03-11T17:59:41Z)
• Mixture of Experts Made Intrinsically Interpretable [34.36996159677674]
  We present textbfMoE-X, a Mixture-of-Experts (MoE) language model designed to be emphintrinsically interpretable. Our approach is motivated by the observation that, in language models, wider networks with sparse activations are more likely to capture interpretable factors. MoE-X achieves perplexity better than GPT-2, with interpretability surpassing even sparse autoencoder (SAE)-based approaches.
  arXiv  Detail & Related papers  (2025-03-05T17:40:54Z)
• Sparse Autoencoder Features for Classifications and Transferability [11.2185030332009]
  We analyze Sparse Autoencoders (SAEs) for interpretable feature extraction from Large Language Models (LLMs)<n>Our framework evaluates (1) model-layer selection and scaling properties, (2) SAE architectural configurations, including width and pooling strategies, and (3) the effect of binarizing continuous SAE activations.
  arXiv  Detail & Related papers  (2025-02-17T02:30:45Z)
• Efficient Redundancy Reduction for Open-Vocabulary Semantic Segmentation [36.46163240168576]
  Open-vocabulary semantic segmentation (OVSS) is an open-world task that aims to assign each pixel within an image to a specific class defined by arbitrary text descriptions.<n>Recent advancements in large-scale vision-language models have demonstrated their open-vocabulary understanding capabilities.<n>This study introduces ERR-Seg, a novel framework that effectively reduces redundancy to balance accuracy and efficiency.
  arXiv  Detail & Related papers  (2025-01-29T13:24:53Z)
• Learning Multiscale Consistency for Self-supervised Electron Microscopy Instance Segmentation [48.267001230607306]
  We propose a pretraining framework that enhances multiscale consistency in EM volumes. Our approach leverages a Siamese network architecture, integrating strong and weak data augmentations. It effectively captures voxel and feature consistency, showing promise for learning transferable representations for EM analysis.
  arXiv  Detail & Related papers  (2023-08-19T05:49:13Z)
• Revisiting the Encoding of Satellite Image Time Series [2.5874041837241304]
  Image Time Series (SITS)temporal learning is complex due to hightemporal resolutions and irregular acquisition times. We develop a novel perspective of SITS processing as a direct set prediction problem, inspired by the recent trend in adopting query-based transformer decoders. We attain new state-of-the-art (SOTA) results on the Satellite PASTIS benchmark dataset.
  arXiv  Detail & Related papers  (2023-05-03T12:44:20Z)
• Learning Efficient Coding of Natural Images with Maximum Manifold Capacity Representations [4.666056064419346]
  The efficient coding hypothesis proposes that the response properties of sensory systems are adapted to the statistics of their inputs. While elegant, information theoretic properties are notoriously difficult to measure in practical settings or to employ as objective functions in optimization. Here we outline the assumptions that allow manifold capacity to be optimized directly, yielding Maximum Manifold Capacity Representations (MMCR)
  arXiv  Detail & Related papers  (2023-03-06T17:26:30Z)
• Unifying Synergies between Self-supervised Learning and Dynamic Computation [53.66628188936682]
  We present a novel perspective on the interplay between SSL and DC paradigms. We show that it is feasible to simultaneously learn a dense and gated sub-network from scratch in a SSL setting. The co-evolution during pre-training of both dense and gated encoder offers a good accuracy-efficiency trade-off.
  arXiv  Detail & Related papers  (2023-01-22T17:12:58Z)
• Task-Oriented Sensing, Computation, and Communication Integration for Multi-Device Edge AI [108.08079323459822]
  This paper studies a new multi-intelligent edge artificial-latency (AI) system, which jointly exploits the AI model split inference and integrated sensing and communication (ISAC) We measure the inference accuracy by adopting an approximate but tractable metric, namely discriminant gain.
  arXiv  Detail & Related papers  (2022-07-03T06:57:07Z)
• Stabilizing Q-learning with Linear Architectures for Provably Efficient Learning [53.17258888552998]
  This work proposes an exploration variant of the basic $Q$-learning protocol with linear function approximation. We show that the performance of the algorithm degrades very gracefully under a novel and more permissive notion of approximation error.
  arXiv  Detail & Related papers  (2022-06-01T23:26:51Z)
• Understanding Self-supervised Learning with Dual Deep Networks [74.92916579635336]
  We propose a novel framework to understand contrastive self-supervised learning (SSL) methods that employ dual pairs of deep ReLU networks. We prove that in each SGD update of SimCLR with various loss functions, the weights at each layer are updated by a emphcovariance operator. To further study what role the covariance operator plays and which features are learned in such a process, we model data generation and augmentation processes through a emphhierarchical latent tree model (HLTM)
  arXiv  Detail & Related papers  (2020-10-01T17:51:49Z)
• Dual-constrained Deep Semi-Supervised Coupled Factorization Network with Enriched Prior [80.5637175255349]
  We propose a new enriched prior based Dual-constrained Deep Semi-Supervised Coupled Factorization Network, called DS2CF-Net. To ex-tract hidden deep features, DS2CF-Net is modeled as a deep-structure and geometrical structure-constrained neural network. Our network can obtain state-of-the-art performance for representation learning and clustering.
  arXiv  Detail & Related papers  (2020-09-08T13:10:21Z)

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.