吴良超的学习笔记

本文主要介绍了如何通过 Keras 实现 LRCN 模型，模型出自论文 Long-term Recurrent Convolutional Networks for Visual Recognition and Description，最近需要用这个模型做个实验，在网上搜到的实现代码不多，因此这里记录一下，以供参考。

This article mainly introduces how to implement the LRCN model through Keras. The model comes from the paper Long-term Recurrent Convolutional Networks for Visual Recognition and Description. I recently needed to use this model for an experiment and didn’t find much implementation code online, so I’m recording it here for reference.

最近需要频繁切换使用台式机和笔记本，但是我的小桌子上实在没法同时放得下一个键盘和笔记本 （≧0≦）。哪怕凑合挤下，还得不停在两台电脑之间切换键鼠，因此就想着有没有共享键鼠的方案，结果在网上找到了 Synergy，试了几天后发现这真的是一个共享键鼠的神器。

This article mainly introduces the principles and implementation of the WAND (Weak AND) algorithm. WAND is a search algorithm applied in scenarios where a query has multiple keywords or tags, and each document also has multiple keywords or tags (such as search engines). Especially when the query contains many keywords or tags, WAND can quickly select the Top n relevant documents. The original paper of the algorithm can be found at Efficient Query Evaluation using a Two-Level Retrieval Process. This article mainly explains the principles of this algorithm and implements it in Python.

本文主要介绍 Wand (Weak And) 算法的原理和实现, Wand 算法是一个搜索算法，应用在 query 有多个关键词或标签，同时每个 document 也有多个关键词或标签的情形（如搜索引擎）；尤其是在 query 中的关键词或标签较多的时候，通过 Wand 能够快速的选择出 Top n 个相关的 document，算法的原始论文见 Efficient Query Evaluation using a Two-Level Retrieval Process，本文主要讲述这个算法的原理以及通过 python 实现这个算法。

This article is a repost, the original link can be found here, authored by carbon zhang. This article mainly introduces several key concepts and classic papers in distributed machine learning, including data parallelism and model parallelism, schools of distributed frameworks, parameter servers, and the evolution of synchronization protocols. It is well worth reading.

文章为转载，原文链接见这里，作者是 carbon zhang。这篇文章主要介绍了分布式机器学习中的若干重点概念和经典论文，包括数据并行和模型并行、分布式框架的流派、参数服务器以及同步协议的演进等，非常值得一看。

When mentioning MapReduce, one naturally thinks of Hadoop MapReduce, but MapReduce is just a programming paradigm, and Hadoop MapReduce is a well-known implementation of this paradigm. Actually, MapReduce can be implemented in various ways. This article introduces how to implement a MapReduce program in Linux bash, with both single-machine and multi-machine versions. The original video is here (requires VPN).

提到 MapReduce，很自然想到的是 Hadoop MapReduce ，但是 MapReduce 只是一个编程范式，而 Hadoop MapReduce 则是这个编程范式的一个比较出名的实现。实际上，可以通过多种方式实现 MapReduce，本文要介绍的就是如何在 Linux 的 bash 下实现一个 MapReduce 程序，并且分别实现了单机版本和多机器版本。原视频见这里，需要自备梯子。

本文主要介绍了互联网几项重要业务（在线广告，推荐系统，搜索引擎）背后所需的一项共同技术：语义理解 (semantic understanding)，同时介绍了实现语义理解的若干种方法：包括矩阵分解，主题模型 (Topic Models) 等。原视频见这里，需要自备梯子。