The paper introduces the concept of internal discrepancy, measuring the gap between the mean representation of positive samples and latent cluster centers in contrastive learning. It derives a generalization bound for downstream supervised tasks that depends on this discrepancy, providing a unifying framework for existing assumptions and motivating a relaxed k‑means loss to reduce it. The study also explores how contrastive learning affects downstream behavior, particularly in class‑imbalanced settings, and establishes properties of the resulting supervised losses.
By Jialiang Wang, Qing Ma, Xiong Zhou, Gangfeng Hu, Haoliang Li, Xianming Liu
arXiv:2608. 11661v1 Announce Type: cross Abstract: A multiplicative dual-encoder network computes a real-valued output for a pair of inputs as the inner product of their separate encodings.
By Zijian Zhao, Sen Li
arXiv:2606. 26617v1 Announce Type: new Abstract: Scaling laws describe how learning performance varies with model size, data size, and compute.
By Ziyan Chen, Zhongzhu Zhou, Ding-Xuan Zhou
The paper investigates the expressive power of multimodal contrastive learning architectures by treating them as parameterized families of joint density estimators. It shows that the classic two‑tower CLIP model is a universal approximator for two modalities, while a common extension that sums pairwise similarities fails to approximate arbitrary joint distributions when three or more modalities are involved, though it can match all pairwise conditionals. To address this limitation, the authors introduce Hadamard‑CLIP, which adds a single learned weight vector to restore universal approximation for any number of modalities while retaining CLIP’s efficient retrieval capabilities.
By Andrew Stuart, Florian Wolf
arXiv:2608. 04234v1 Announce Type: cross Abstract: We study the problem of aligning data from multiple modalities into a shared representation space, focusing on settings where strong pretrained unimodal encoders are available but cross-modal paired data are scarce.
By Yixuan Florence Wu, Yilun Zhu, Naichen Shi
arXiv:2606. 04280v1 Announce Type: cross Abstract: Contrastive learning has become a leading paradigm for self-supervised representation learning, yet the conditions under which it recovers meaningful latent geometry remain incompletely understood.
By Justinas Zaliaduonis, Patrick Putzky, Till Richter, Sergios Gatidis
arXiv:2602. 24012v2 Announce Type: replace Abstract: Contrastive learning has become a cornerstone of modern representation learning, allowing training with massive unlabeled data for both task-specific and general (foundation) models.
By Roy Betser, Eyal Gofer, Meir Yossef Levi, Guy Gilboa
MoEMB introduces a mixture‑of‑experts (MoE) approach to scale universal multimodal embeddings (UME) without increasing the size of the output vector or relying on autoregressive decoding. By expanding encoder capacity along the expert axis, MoEMB achieves state‑of‑the‑art performance on MMEB‑V2 and MRMR benchmarks with only 3 B active parameters, outperforming TTE‑based methods that use more than four times as many active parameters and require significantly more compute. The paper also presents the first comprehensive study of adaptive computation for MoE‑based embeddings, exploring training‑time and inference‑time strategies to further improve efficiency for large‑scale retrieval and recommendation systems.
By Xuanming Cui, Shlok Kumar Mishra, Wentao Bao, Aashu Singh, Zihao Wang, Xiangjun Fan, Jun Xiao, Ser-Nam Lim, Jianpeng Cheng
The paper introduces a divergence-based similarity function (DSF) for multi-view contrastive learning, representing each set of augmented views as a distribution and measuring similarity via distribution divergence. DSF captures joint structure across all views, outperforming prior pairwise methods on tasks such as kNN classification, linear evaluation, transfer learning, and distribution shift. It also offers greater efficiency and eliminates the need for a temperature hyperparameter, unlike cosine similarity.
By Jaehyoung Jeon, Cheolsu Lim, Myungjoo Kang
InfMasking introduces a unified multimodal learning framework that uses Infinite Masking to encourage fusion modules to capture synergistic information across modalities. By randomly masking modality features before fusion and enforcing consistency with an unmasked reference, the method discourages reliance on dominant or shared signals and promotes meaningful cross‑modal interactions. The approach models masked fused representations as Gaussian, optimizing closed‑form variational bounds, and demonstrates improved predictive performance and higher synergy metrics on benchmark datasets.
By Liangjian Wen, Qun Dai, Jiang Duan, Yong Dai, Jianzhuang Liu, Dongkai Wang, Jun Wang, Fuyuan Cao, Zenglin Xu, Zhao Kang
arXiv:2603.02767v4 Announce Type: replace-cross
Abstract: Image--text contrastive pretraining has become a dominant paradigm for visual representation learning, yet existing methods often yield repre...
By Hanpeng Liu, Zidan Wang, Shuoxi Zhang, Zonglin Zhao, Zihao Bo, Rinyoichi Takezoe, Kaiwen Long, Yaqian Li, Kun He
arXiv:2607. 16295v1 Announce Type: cross Abstract: Mechanistic interpretability has made significant strides in understanding neural network representations, with sparse dictionary learning (SDL) methods, most prominently sparse autoencoders, as a central paradigm.
By Yiming Tang, Qinglin Qi, Zhaoqian Yao, Harshvardhan Saini, Dianbo Liu