MLPerf Tiny
收藏资源简介:
MLPerf Tiny是由哈佛大学等超过50个学术和工业组织合作开发的首个行业标准基准套件,专为超低功耗微型机器学习系统设计。该数据集包含四个基准:关键词识别、视觉唤醒词、图像分类和异常检测,旨在评估机器学习推理的准确性、延迟和能耗,以公平可重复的方式展示不同系统间的权衡。数据集的应用领域广泛,从智能门铃到工业异常检测,旨在推动微型机器学习技术的发展和标准化。
MLPerf Tiny is the first industry-standard benchmark suite developed in collaboration with over 50 academic and industrial organizations including Harvard University, specifically designed for ultra-low-power tiny machine learning systems. This dataset includes four benchmarks: keyword spotting, visual wake words, image classification, and anomaly detection, which aim to evaluate the accuracy, latency, and energy consumption of machine learning inference and demonstrate the trade-offs between different systems in a fair and reproducible manner. The dataset covers a wide range of application scenarios, from smart doorbells to industrial anomaly detection, and is intended to promote the development and standardization of tiny machine learning technologies.




