知识卡片
移动推荐同时受资源预算约束
移动端推理需要同时考虑模型大小、运行时间与存储,不能只按服务器端质量选择结构。压缩嵌入和利用较强模型训练较小模型可以降低部署开销,但必须在目标设备与任务上重新验证效果,不能把压缩比例当作真实加速。
这张卡片看不懂?
这张卡片已通过人工审查,不再接受失败反馈。
已有反馈 · 0 条
还没有人提交反馈。
在整书笔记中的位置
原书出处 · 1 处
- · 查看原书依据
LLRec adapts to the resource-constrained nature of mobile devices. Here tensor-train decomposition is used that compresses parameters in the embedding layer. As per the authors, this model has less time and memory consumption. The factors chosen by the authors are the user, POI, and a timestamp for check-in. Here check-in sequence with consecutive activities by the user within the period has been considered. The contextual
参考来源
- 位置:《Artificial Intelligence and Data Science in Recommendation System: Current Trends, Technologies, and Applications》第09章《兴趣点推荐的影响参数》(来源项目:a19c7933-4eda-4176-aebd-30fe7e1ad292;源文件:extracted/Artificial Intelligence and Data Science in Recommendation System: Current Trends, Technologies, and Applications/_epub-src/OEBPS/Text/9789815136746-C9.html)
- 结论依据:摘录中关于移动推荐同时受资源预算约束的定义、比较或条件支撑卡片;原书表述不严谨的范围已在覆盖记录逐项限定,不把案例结果外推为普遍结论。
- 原始内容:LLRec adapts to the resource-constrained nature of mobile devices. Here tensor-train decomposition is used that compresses parameters in the embedding layer. As per the authors, this model has less time and memory consumption. The factors chosen by the authors are the user, POI, and a timestamp for check-in. Here check-in sequence with consecutive activities by the user within the period has been considered. The contextual