Frozen self-supervised representations often transfer well with only a few labels across many semantic tasks. We argue that a single geometric quantity, \emph{directional} CDNV (decision-axis variance), sits at the core of two favorable behaviors: strong few-shot transfer within a task, and low interference across many tasks. We show that both emerge when variability \emph{along} class-separating directions is small. First, we prove sharp non-asymptotic multiclass generalization bounds for downstream classification whose leading term is the directional CDNV. The bounds include finite-shot corrections that cleanly separate intrinsic decision-axis variability from centroid-estimation error. Second, we link decision-axis collapse to multitask geometry: for independent balanced labelings, small directional CDNV across tasks forces the corresponding decision axes to be nearly orthogonal, helping a single representation support many tasks with minimal interference. Empirically, across SSL objectives, directional CDNV collapses during pretraining even when classical CDNV remains large, and our bounds closely track few-shot error at practical shot sizes. Additionally, on synthetic multitask data, we verify that SSL learns representations whose induced decision axes are nearly orthogonal. The code and project page of the paper are available at [\href{https://dlfundamentals.github.io/directional-neural-collapse/}{project page}].
On Leveraging Variational Graph Embeddings for Open World Compositional Zero-Shot Learning
Technische Universität München(慕尼黑工业大学);Mercateo AG(Mercateo股份公司)
人类能够识别和分类已知概念的新组合。组合零镜头学习(Compositional Zero-Shot learning,CNOML)的任务是学习原始概念(即对象和状态)的组合,即使是它们的新组合也可以被zero-shot分类。在这项工作中,我们不假设任何先验知识的可行性,即开放世界的设置,其中不可行的组合物占主导地位的搜索空间的新组合物。我们提出了一种组合变分图自动编码器(CVGAE)的方法来学习原始概念(节点)的变分嵌入以及它们的组合(通过边缘)的可行性。这样的建模使得CVGAE可扩展到现实世界的应用场景。这与SOTA方法CGE相反,SOTA方法CGE在计算上非常昂贵。例如,对于基准C-GQA数据集,CGE需要3.94 x 10^5个节点,而CVGAE只需要1323个节点。我们学习将图和图像嵌入映射到公共嵌入空间。CVGAE采用深度度量学习方法,并通过投影图和图像嵌入之间的双向对比损失来学习该空间中的相似性度量。我们在三个基准数据集上验证了该方法的有效性,并通过一个图像检索任务证明了CVGAE学习的表示更适合于组合泛化。
HEALing Entropy Collapse: Enhancing Exploration in Few-Shot RLVR via Hybrid-Domain Entropy Dynamics Alignment
School of Informatics, Xiamen University(厦门大学信息学院);Tsinghua University(清华大学);Xiamen Unisound Intelligence Technology Co., Ltd(厦门Unisound智能科技有限公司);Key Laboratory of Digital Protection and Intelligent Processing of Intangible Cultural Heritage of Fujian and Taiwan (Xiamen University), Ministry of Culture and Tourism, China(福建省和台湾非物质文化遗产数字化保护与智能处理重点实验室(厦门大学),中华人民共和国文化和旅游部,中国)
From Coefficients to Directions: Rethinking Model Merging with Directional Alignment
University of Oxford(牛津大学);Tsinghua University(清华大学);Nanyang Technological University(南洋理工大学);East China Normal University(华东师范大学);RIKEN(日本研究机构);The University of Tokyo(东京大学);Southern University of Science and Technology(南方科技大学)
Spectral Propagation Graph Network for Few-shot Time Series Classification
Institute of Medical Technology, Health Science Center of Peking University(北京大学医学部医学技术研究院);National Institute of Health Data Science, Peking University(北京大学国家健康医疗大数据研究院)
Latent Iterative Refinement Flow: A Geometric Constrained Approach for Few-Shot Generation
School of Mathematics(数学学院);Statistics, Huazhong University of Science(统计学,华中科技大学);School of Electronic Information(电子信息学院);Communications, Huazhong University of Science(通信学院,华中科技大学);Center for Mathematical Science, Huazhong University of Science(数学科学中心,华中科技大学);Huawei(华为)
A Markovian View of Iterative-Feedback Loops in Image Generative Models: Neural Resonance and Model Collapse
Luddy School of Informatics, Computing, and Engineering, Indiana University, Bloomington, IN, USA(印第安纳大学信息学、计算与工程学院);Department of Computer Science, Stonehill College, Easton, MA, USA(石溪学院计算机科学系)
RadioDiff-FS: Physics-Informed Manifold Alignment in Few-Shot Diffusion Models for High-Fidelity Radio Map Construction
State Key Laboratory of ISN and School of Telecommunications Engineering, Xidian University(信息与通信工程国家重点实验室和西安电子科技大学电信工程学院);School of Physics, Xidian University(西安电子科技大学物理学院);Department of Electrical and Computer Engineering, University of Waterloo(滑铁卢大学电气与计算机工程系)
Finite-Sample Metric Non-Collapse for Geometrically Supervised Latent World Models in Control
Naveen Jindal School of Management University of Texas at Dallas, Richardson, TX, 75080 USA;Department of Mathematics Texas A\&M University, College Station, TX, 77843 USA;Naveen Jindal School of Management University of Texas at Dallas, Richardson, TX 75080, USA;Department of Mathematics Texas A\&M University, College Station, TX 77843, USA
Neural Collapse Terminus: A Unified Solution for Class Incremental Learning and Its Variants
King Abdullah University of Science and Technology(阿卜杜拉国王科技大学);Wuhan University(武汉大学);Peking University(北京大学);Harbin Institute of Technology (Shenzhen)(哈尔滨工业大学(深圳));University of Oxford(牛津大学);University of Sydney(悉尼大学)