RamanBench
收藏资源简介:
RamanBench是首个面向拉曼光谱机器学习的大规模可复现基准测试,由柏林工业大学等机构联合创建。该基准整合了74个跨四领域的数据集(含16个新发布数据集),涵盖325,668个光谱,支持分类和回归任务。数据集通过标准化API提供访问,来源包括HuggingFace、Kaggle等8个平台,覆盖材料科学、生物技术等应用场景。其创建过程严格遵循可学习性和最小规模标准,旨在解决拉曼光谱领域数据碎片化、评估不一致的问题,推动医疗诊断、生物研究等关键应用的算法发展。
RamanBench is the first large-scale reproducible benchmark for machine learning on Raman spectroscopy, jointly created by institutions including Technische Universität Berlin and others. This benchmark integrates 74 datasets spanning four domains (including 16 newly released datasets), covering 325,668 Raman spectra, and supports both classification and regression tasks. The datasets are accessible via a standardized API, sourced from 8 platforms including HuggingFace and Kaggle, covering application scenarios such as materials science and biotechnology. Its development strictly follows the criteria of learnability and minimal scale, aiming to address the issues of data fragmentation and inconsistent evaluation in the field of Raman spectroscopy, and promote the advancement of algorithms for key applications including medical diagnosis and biological research.

- 1RamanBench: A Large-Scale Benchmark for Machine Learning on Raman Spectroscopy柏林工业大学; 柏林应用科技大学; KWS SAAT; 下莱茵应用技术大学; VTT芬兰; 柏林工程应用技术大学; 爱因斯坦数字未来中心 · 2026年



