Uniform Manifold Approximation and Projection
PulseAugur coverage of Uniform Manifold Approximation and Projection — every cluster mentioning Uniform Manifold Approximation and Projection across labs, papers, and developer communities, ranked by signal.
2 天有情绪数据
-
新方法解决降维中的模糊数据点
研究人员引入了一种新颖的基于图的方法来解决降维中的模糊实例，这是视觉伪影的常见来源。该方法识别出在高维空间中与多个不同邻域高度相似的数据点。通过将这些模糊实例复制为投影中的多个点，并将每个点放置在其相关的邻域内，该方法旨在更准确地表示数据的结构并减少局部邻域嵌入。
-
ML classifier automates refactoring of BDD test suites
Researchers have developed a method to automatically identify and categorize opportunities for refactoring in behavior-driven development (BDD) software test suites. Their approach uses machine learning classifiers, spe…
-
New framework detects manipulative political narratives on social media
Researchers have developed a new framework to detect and categorize manipulative political narratives found on social media. The system first uses a few-shot prompt with a reasoning model to filter out manipulative post…
-
LLM助力城市桥梁重要性评分与图分析聚类
研究人员开发了一种使用异构图分析和大型语言模型评估城市桥梁重要性的新方法。该方法基于来自开放数据源的交通通达性、医院邻近性和供应链影响等因素量化桥梁的重要性。然后，系统利用聚类技术识别不同城市的功能性桥梁原型，并采用LLM（特别是Elyza8b）自动生成与政策相关的解释。
-
UMAP降维方法与PCA和t-SNE的比较
一篇新论文将均匀流形逼近与投影（UMAP）与其他降维技术（如PCA和t-SNE）进行了比较。该研究使用模拟和真实世界的数据集，系统地评估了监督式UMAP在回归和分类任务中的应用。结果表明，虽然监督式UMAP在分类任务中有效，但在回归任务中难以整合响应信息，这表明了未来研究的方向。
-
New Class Angular Distortion Index metric improves dimensionality reduction faithfulness
Researchers have introduced the Class Angular Distortion Index (CADI), a novel metric for evaluating dimensionality reduction techniques. CADI addresses limitations in existing metrics by assessing the faithfulness of c…
-
VERA tool automatically explains 2D data embeddings with region annotations
Researchers have developed VERA, a new method for automatically generating visual explanations of two-dimensional data embeddings. VERA identifies key regions within these embeddings and links them to human-interpretabl…
-
扩散Transformer生成合成欺诈数据以改进检测
研究人员开发了一种名为EmDT的新扩散模型，旨在为欺诈检测生成合成数据。该模型利用UMAP聚类来识别特定的欺诈模式，并利用Transformer网络在数据生成过程中捕获特征关系。在信用卡欺诈数据集上的实验表明，与现有方法相比，EmDT显著提高了XGBoost等下游分类器的性能。
-
Topology tool Mapper reveals how language models encode ambiguity
Researchers have introduced Mapper, a topological data analysis tool, to better understand how language models handle ambiguity. Applied to RoBERTa-Large, Mapper revealed that fine-tuning reorganizes the model's embeddi…
-
UMAP dimensionality reduction forces analyzed for cluster formation
This paper delves into the mechanics of Uniform Manifold Approximation and Projection (UMAP), a popular dimensionality reduction technique. Researchers analyzed the attractive and repulsive forces UMAP uses to map high-…
-
Manifold learning accurately detects cardiac arrhythmias without labels
Researchers have demonstrated the effectiveness of nonlinear dimensionality reduction (NLDR) algorithms, such as UMAP and t-SNE, for unsupervised detection of cardiac arrhythmias from electrocardiogram (ECG) signals. Un…
-
新研究质疑平坦最小值，提出拓扑忠实降维方法
研究人员开发了 DiRe-RAPIDS，一种新的降维技术，与 UMAP 和 t-SNE 等现有方法相比，它能更好地保留高维数据的全局拓扑结构。DiRe-RAPIDS 在一个旨在评估噪声流形拓扑忠实度的新基准上进行了调优。在一个大型的 arXiv 论文嵌入数据集上，DiRe-RAPIDS 在可比的速度下，比 UMAP 保持了显著更多的拓扑结构。此外，还引入了一个新的框架，用于定量和可视化分析参数化投影方法的局部邻域不稳定性，并展示了其在…