Tag
该论文提出 SAGA 方法，将任务边界处训练的稀疏自编码器转换为对表示漂移的低秩局部约束，以缓解持续学习中的灾难性遗忘，并在三个 ViT-B/16 基准上取得最佳的 Class-IL 准确率与最低遗忘率。
提出“隐喻电路”假说，解释 Vision Transformer 如何在缺乏直接指称证据时通过具体锚点概念（如 fire）间接接地抽象概念（如 angry），并利用 Transcoders 对 CLIP 和 DINO 的电路追踪与因果干预进行了验证。
该论文提出 reViT，通过在每个递归深度用一个共享专家库的凸组合来表示 FFN，使单个 Transformer 块以递归方式应用即可在相近推理 FLOPs 下匹配全深度视觉编码器的精度，并支持弹性深度部署与权重量级压缩。
该论文通过冻结选中子集并只更换训练学习器的受控实验，证明样本价值（以及 easy-first 与几何覆盖策略的 crossover budget）是相对于目标学习器定义的，而非数据的固有属性。
This paper introduces randomized-pass replay (RPR) to bound rehearsal gaps in online continual learning, showing improved accuracy over independent class-balanced retrieval in experience replay methods like ER-ACE.
This paper presents a parameter-efficient adaptation framework using LoRA and contrastive learning to enhance vision transformers for automatic target recognition in synthetic aperture sonar imagery.
The author revisits influential open-source works in representation learning, listing key papers from MoCo v1 to LeJEPA that advanced vision foundation models and self-supervised learning.
Proposes a decoupled training strategy that adapts normalization layers and uses precomputed features to reduce overhead in transfer learning, achieving competitive accuracy with significantly reduced training time and energy consumption.
This paper explores the use of register tokens in pixel-space Diffusion Transformers (DiTs), finding they improve feature map quality despite DiTs lacking patch-token outliers. The authors propose Register Guidance, a technique to amplify register contributions for better visual structure.
UniverSat introduces a Universal Patch Encoder for Vision Transformers that enables robust, sensor-agnostic spatial feature extraction across diverse Earth Observation data types, achieving strong results on classification and segmentation benchmarks.
ViT-Up introduces a task-agnostic feature upsampler for Vision Transformers that predicts features at arbitrary continuous image coordinates, enabling dense feature maps at any resolution and improving dense prediction and semantic correspondence benchmarks. It outperforms prior state-of-the-art upsamplers, with gains of up to +2.07 mIoU on Cityscapes and +4.17 [email protected] on SPair-71k.
Phase Marginalization is a post-hoc method that addresses phase-dependent instability in Vision Transformers by evaluating structured patch-grid phases and aggregating outputs. It improves segmentation, depth, and local matching over the canonical baseline with minimal extra cost.
This article presents a new paper on Elastic Attention Cores for Vision Transformers, proposing a core-periphery block-sparse attention structure that improves scalability and accuracy compared to dense self-attention methods like DINOv3.