Structuring Sparsity: Block-Sparse Featurizers Capture Visual Concept Manifolds

• URL: http://arxiv.org/abs/2606.25234v1
• Date: Tue, 23 Jun 2026 23:28:30 GMT
• Title: Structuring Sparsity: Block-Sparse Featurizers Capture Visual Concept Manifolds
• Abstract summary: We show that concepts are often realized as geometric structures in low dimensional regions of activation space.<n>We implement three variants of block-sparse featurizers (BSFs) and, through a minimum-description-length analysis, show that all three describe activations more compactly than direction-based featurizers.
• Score: 40.072358858272445
• License: http://creativecommons.org/licenses/by/4.0/
• Abstract: What is the geometry of a visual percept? The most widely used protocols for decomposing neural network representations into interpretable parts treat concepts as isolated directions, yet recent work shows that concepts are often realized as geometric structures in low dimensional regions of activation space. We turn to the literature of Structured sparsity to close this gap, and show that block sparsity, which groups directions into blocks, is the prior matched to a generative model in which a representation is a sparse sum of low-dimensional manifolds: the modern, learned form of a classical idea in visual neuroscience, where a visual feature is carried by a coordinated group of neurons rather than a single tuned one. We implement three variants of block-sparse featurizers (BSFs) and, through a minimum-description-length analysis, show that all three describe activations more compactly than direction-based featurizers, with the recovered concepts typically two- to four-dimensional. We then use BSFs to (i) recontextualize prior work, showing that curve detectors in InceptionV1 actually read from a single continuous curve manifold, (ii) discover novel manifolds including shadows and lighting in DINOv3, and (iii) support interpretable control of image generation in diffusion models (SDXL) via manifold steering.

Related papers

• OpenGaFF: Open-Vocabulary Gaussian Feature Field with Codebook Attention [80.51557267896938]
  We present OpenGaFF, a novel framework for open-vocabulary 3D scene understanding built upon 3D Gaussian Splatting.<n>At the core of our method is a Gaussian Feature Field that models semantics as a continuous function of Gaussian geometry and appearance.<n>To further enforce object-level semantic consistency, we introduce a structured codebook that serves as a set of shared semantic primitives.
  arXiv  Detail & Related papers  (2026-05-07T12:10:07Z)
• Learning 3D Representations for Spatial Intelligence from Unposed Multi-View Images [81.94999489820974]
  UniSplat is a feed-forward framework for learning 3D representations from unposed multi-view images.<n>We introduce a dual-masking strategy that strengthens geometry induction in the encoder.<n>Second, we develop a coarse-to-fine Gaussian splatting strategy that reduces appearance-semantics inconsistencies.<n>Third, we introduce a pose-conditioned recalibration mechanism that interrelates predicted 3D point and semantic maps into the image plane.
  arXiv  Detail & Related papers  (2026-04-12T10:36:18Z)
• FF3R: Feedforward Feature 3D Reconstruction from Unconstrained views [34.14517479047999]
  This paper introduces FF3R, a fully annotation-free feed-forward framework that unifies geometric and semantic reasoning from unconstrained multi-view image sequences.<n>Unlike previous methods, FF3R does not require camera poses, depth maps, or semantic labels, relying solely on rendering supervision for RGB and feature maps.<n>Two key innovations: (i) a Token-wise Fusion Module that enriches geometry tokens with semantic context via cross-attention, and (ii) a Semantic-Geometry Mutual Boosting mechanism combining geometry-guided feature warping for global consistency with semantic-aware voxelization for local coherence.
  arXiv  Detail & Related papers  (2026-04-10T19:45:24Z)
• O3N: Omnidirectional Open-Vocabulary Occupancy Prediction [31.91030387170798]
  We present O3N, the first purely visual, end-to-end Omnidirectional Open-vocabulary Occupancy predictioN framework.<n>O3N embeds omnidirectional voxels in a polar-spiral topology, enabling continuous spatial representation and long-range context modeling.<n>Our method achieves state-of-the-art performance on QuadOcc and Human360Occ benchmarks.
  arXiv  Detail & Related papers  (2026-03-12T16:45:42Z)
• CVP: Central-Peripheral Vision-Inspired Multimodal Model for Spatial Reasoning [48.36177110428022]
  We present a central-peripheral vision-inspired framework (CVP) for spatial reasoning.<n>CVP draws inspiration from the two types of human visual fields -- central vision and peripheral vision.<n> Experiments show that CVP achieves state-of-the-art performance across a range of 3D scene understanding benchmarks.
  arXiv  Detail & Related papers  (2025-12-09T00:21:13Z)
• Into the Rabbit Hull: From Task-Relevant Concepts in DINO to Minkowski Geometry [31.26429968473424]
  DINOv2 is routinely deployed to recognize objects, scenes, and actions; yet the nature of what it perceives remains unknown.<n>As a working baseline, we adopt the Linear Representation Hypothesis (LRH) and operationalize it using SAEs.<n>We produce a 32,000-unit dictionary that serves as the interpretability backbone of our study.
  arXiv  Detail & Related papers  (2025-10-08T22:42:20Z)
• Cross-Layer Discrete Concept Discovery for Interpreting Language Models [13.842670153893977]
  Cross-layer VQ-VAE is a framework that uses vector quantization to map representations across layers.<n>Our approach uniquely combines top-k temperature-based sampling during quantization with EMA codebook updates.
  arXiv  Detail & Related papers  (2025-06-24T22:43:36Z)
• Concept-Guided Interpretability via Neural Chunking [64.6429903327095]
  We show that neural networks exhibit patterns in their raw population activity that mirror regularities in the training data.<n>We propose three methods to extract recurring chunks on a neural population level.<n>Our work points to a new direction for interpretability, one that harnesses both cognitive principles and the structure of naturalistic data.
  arXiv  Detail & Related papers  (2025-05-16T13:49:43Z)
• GOV-NeSF: Generalizable Open-Vocabulary Neural Semantic Fields [50.68719394443926]
  Generalizable Open-Vocabulary Neural Semantic Fields (GOV-NeSF) is a novel approach offering a generalizable implicit representation of 3D scenes with open-vocabulary semantics. GOV-NeSF exhibits state-of-the-art performance in both 2D and 3D open-vocabulary semantic segmentation.
  arXiv  Detail & Related papers  (2024-04-01T05:19:50Z)
• A Self-supervised Mixed-curvature Graph Neural Network [76.3790248465522]
  We present a novel Self-supervised Mixed-curvature Graph Neural Network (SelfMGNN) We show that SelfMGNN captures the complicated graph structures in reality and outperforms state-of-the-art baselines.
  arXiv  Detail & Related papers  (2021-12-10T08:56:55Z)

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.