《Merlin NVTabular:基于 GPU 加速的推荐系统特征工程最佳实践.pdf》由会员分享,可在线阅读,更多相关《Merlin NVTabular:基于 GPU 加速的推荐系统特征工程最佳实践.pdf(29页珍藏版)》请在三个皮匠报告上搜索。
1、Merlin NVTabular: 基于GPU加速 的推荐系统特征工程最佳实践 黄孟迪,NVIDIA深度学习工程师 2 RELATED SESSIONS IN GTC CHINA Merlin:GPU 加速的推荐系统框架 CNS20590 - 王泽寰, 英伟达亚太AI开发者技术经理, NVIDIA Merlin HugeCTR:深入研究性能优化 CNS20516 Minseok Lee, GPU计算专家, NVIDIA Merlin NVTabular: 基于GPU加速的推荐系统特征工程最佳实践 CNS20624 黄孟迪,深度学习 工程师,NVIDIA GPU加速的数据处理在推荐系统中的应用
2、CNS20813 - 魏英灿,GPU计算专家, NVIDIA 将HugeCTR Embedding集成于TensorFlow CNS20377 - 董建兵,GPU计算专家, NVIDIA 使用GPU embedding cache加速CTR推理过程 CNS20626 郁凡, GPU计算专家, NVIDIA Learning More About NVIDIA Merlin Merlin Overview NVTabular - Merlin ETL Tutorials - Best Practices For RecSys Feature Engineering Goal 1: Improvi
3、ng Model Accuracy Goal 2: Quick experimentation with GPU Acceleration Goal 3: Scale to Production Systems With NVTabular Agenda Merlin Overview 5 Industrial Recommendation Challenges Multiple iterations can consume a lot of time to find the most accurate feature set Tabular data scales poorly using the common deep learning method of item by item Large embedding tables require significant memory an