Software

MXNet: Optimizing Memory Consumption in Deep Learning

[MXNet: Optimizing Memory Consumption in Deep Learning](https://mxnet.incubator.apache.org/versions/master/architecture/note_memory.html)

Software

MXNet: Dependency Engine for Deep Learning

[MXNet: Dependency Engine for Deep Learning](https://mxnet.incubator.apache.org/versions/master/architecture/note_engine.html)

分布式平台

pyspark中的HBaseConverters

最近项目上有个需求，使用 pyspark 读取 HBase 中存储的 java.math.BigDecimal。

论文阅读笔记

Revisiting Spatial-Temporal Similarity: A Deep Learning Framework for Traffic Prediction

AAAI 2019。网格流量预测，两个问题：空间依赖动态性，另一个是周期平移。原文链接：[Revisiting Spatial-Temporal Similarity: A Deep Learning Framework for Traffic Prediction](http://export.arxiv.org/abs/1803.01254)

论文阅读笔记

Deep Spatio-Temporal Residual Networks for Citywide Crowd Flows Prediction

AAAI 2017, ST-ResNet，网格流量预测，用三个相同结构的残差卷积神经网络对近邻时间、周期、趋势（远期）分别建模。与 RNN 相比，RNN 无法处理序列长度过大的序列。三组件的输出结果进行集成，然后和外部因素集成，得到预测结果。原文地址：[Deep Spatio-Temporal Residual Networks for Citywide Crowd Flows Prediction](https://arxiv.org/abs/1610.00081)