小鱼日记

From Reusing to Forecasting: Accelerating Diffusion Models with TaylorSeersMeta Author：Shanghai Jiao Tong University Conference：ICCV 2025 Link：TaylorSeers Motivation “cache-then-reuse” :远距离步骤复用特征会显著损害生成质量 related work:acceleration techniques: Sampling Timestep Reduction Denoising Network Acceleration: Model Compression-based Feature Caching-based Method Cache-then-forecast Paradigm： 对特征进行时序建模实现对扩散模型未来时间步特征的预测 Framework 泰勒公式： results in text-to-video Note Consistency...

Meta Title：Multi-view Spectral Clustering Network Author：Sichuan University Conference：IJCAI Year：2019 Link：MvSCN MainIntro the key to discriminative MvC is formulating the within-view similarity and the between-view consistency Contribution: learning the local invariance by SiameseNet construct an orthogonal layer and propose an optimization method the proposed MvSCN could be the first deep extension of multi-view spectral clustering Background Spectral Clustering Multi-view...

论文阅读记录Constitutional AI 标题：Constitutional AI: Harmlessness from AI Feedback 作者：Anthropic 年份：2022 链接：Cai Main Framework: Motivation: Scaling Supervision: leverage AI to help humans to more efficiently supervise AI AI supervision may be more efficient than collecting human feedback AI systems can already perform some tasks at or beyond human level Visualization: A Harmless but Non-Evasive (Still Helpful) Assistant Simplicity and Transparency The Constitutional AI Approach: human...

NoteRALI 标题：REASONING AS REPRESENTATION: RETHINKING VISUAL REINFORCEMENT LEARNING IN IMAGE QUALITY ASSESSMENT 作者：Bytedance & PKU 发表会议/期刊：ICLR 年份：2026 链接：RALI Intro Topic: this paper focuses on the source of generalization of RL-based IQA models (e.g., Q-Insight) Motivation: 引入视觉强化学习的IQA模型，其泛化能力提升背后的原理缺乏系统性分析（尽管已有研究在其他领域探索了强化学习的泛化，但图像质量评估任务中视觉特征的独特复杂性和质量评估的主观性，使得这些发现难以直接迁移） 逐步推理会带来高延迟和高加载开销，限制了在在线强化学习、移动端和实时场景中的部署 two critical questions(RACT & RALI): How is generalization...

论文阅读记录MAVEN-ARG 标题：MAVEN-ARG: Completing the Puzzle of All-in-One Event Understanding Dataset with Event Argument Annotation 作者：Xiaozhi Wang(THU) 发表会议/期刊：ACL 年份：2024 链接：MAVEN Intro Background: event understanding is typically organized as three information extraction tasks: event detection (ED), event argument extraction (EAE), event relation extraction (ERE). MAVEN MAVEN-ERE MAVEN-ARG Motivation：A large-scale dataset covering all the event understanding tasks has long been...

论文阅读记录stMMR 标题：stMMR: accurate and robust saptial domain identification from spatially resolved transcriptomics with multimodal feature representation 作者：Shandong University 发表会议/期刊：GigaScience 年份：2024 链接：stMMR 主要内容简介 Motivation:融合多模态困难，多模态数据间具有显著异质性，而且在数据尺度和分辨率上存在差异 方法与创新点 总体框架：3 steps: multimodal feature embedding, feature fusion, and feature reconstruction 方法概述： Multimodal feature embedding gene expression: $ G \in \mathbb{R}^{n \times p} $ (n spots, p genes) histological...

论文阅读记录Bering 标题：Bering: joint cell segmentation and annotation for spatial transcriptomics with transferred graph embeddings 作者：Harvard University 发表会议/期刊：Nature Communications 年份：2025 链接：Bering 主要内容简介 Motivation:Some tissues have densely packed cells with unclear boundaries, making it difficult to perform accurate segmentation Task:Cell segmentation and annotation for spatial transcriptomics 方法与创新点 总体框架： 方法概述： 图构建，NGC 图卷积和全连接网络 节点分类 边嵌入由三部分组成- node representation-...

论文阅读记录IE-HERCL 标题：Image-Enhanced Hybrid Encoding with Reinforced Contrastive Learning for Spatial Domain Identification in Spatial Transcriptomics 作者：Central South University 发表会议/期刊：IJCAI 年份：2025 链接：IE-HERCL 主要内容简介 Motivation: Exisisting methods fail to account for complex interdependencies between modalities. 方法与创新点 总体框架： 方法概述： Multimodal Feature Representation Learning utilize AE to extract & Loss function L_rec: $$L^g_{rec} = | X_{g,i} -...

论文阅读记录stLearn 标题：Robust mapping of spatiotemporal trajectories and cell–cell interactions in healthy and diseased tissues 作者：The University of Queensland 发表会议/期刊：Nature Communications 年份：2023 链接：stLearn 主要内容简介 Motivation: the (re)construction of spatio-temporal trajectories the study of cell–cell interactions the improvement of spatial data quality by imputation 方法与创新点 总体框架： 方法概述： pseudo-time-space (PSTS) for spatio-temporal trajectory inference DPT 与 Space...

论文阅读记录GraphST 标题：Spatially informed clustering, integration,and deconvolution of spatial transcriptomics with GraphST 作者：National University of Singapore (NUS) 发表会议/期刊：Nature Communications 年份：2023 链接：GraphST 主要内容简介 Motivation: 反卷积未利用空间信息 多样本整合未利用空间信息 方法与创新点 总体框架： 方法概述： graph self-supervised contrastive learning framework： data augmentation GNN-based encoder for representation learning self-supervised contrastive learning for representation refinement ...