Liu Sida's Homepage

Machine Learning & Human Learning

© 2014. liusida All rights reserved.

Nuance in Monty Hall Paradox

Marilyn vos Savant has made a mistake. She knew the game show Let’s Make a Deal too well, that she assumed the rules of the game show also applied to the question she was asked.

On the websi...

What is Mathematics According to Keith Devlin

上午读了Keith Devlin教授的课程背景材料,有点感触,摘抄了几段对我很有启发的文字,以备日后参考。

In Keith Devlin’s book ‘Introduction to Mathematical Thinking’, he writes,

“Virtually nothing (with just two further advances, bo...

Dynamic NN Allowing Additional Evidence?

We have traditional Neural Network (NN), with static structure like this:

signal -> input -> hidden layer -> prediction =?= truth

-> means to propagate forward =?= means ...

Why does the person with highest IQ not become the most successful one?

I heard about the “Study of Mathematically Precocious Youth After 35 Years” years ago, but after studying machine learning, especially the generalization problem, I guess I have glanced some possib...

Use Tensorflow to Compute Gradient

In most of Tensorflow tutorials, we use minimize(loss) to automatically update parameters of the model.

In fact, minimize() is an integration of two steps: computing gradients, and applying ...

Scree of PCA(Principal Component Analysis)

I learned the concept of PCA today, and found out this method of reducing dimension is quite terse.

If we do PCA to a 40-d dataset, reduce it into a 2-d dataset, it simply choose the 2 most ...

Learning Rate is Too Large

What if I see a training accuracy scalar graphic like this:

The accuracy curve of training mini-batch is going down a little bit over time after reached a relative high point. That m...

Manipulating Tensorboard

Tensorboard is a very useful tool for visualizing the logs of Tensorflow. It is now an independent project on GitHub, here’s the link.

In the past, if we were doing small projects, we usuall...

Implement a Deep Neural Network using Python and Numpy

I have just finished Andrew Ng’s new Coursera courses of Deep Learning (1-3). They are one part of his new project

In those courses, there is a series of interview of Heroes...

Cross Entropy 的通俗意义

cross_entropy 公式如下:

它描述的是可能性 S 到 L 的距离,也可以说是描述用 S 来描述 L 还需要多少信息(如果是以2为底的log,则代表还需要多少bit的信息;如果是以10为底的log,则代表还需要多少位十进制数的信息)。

当年 香农 Shannon 创立信息论的时候,考虑的是每一次都是扔硬币,结果只有2个可能,所以用的是以2为底,发明了...


今天学习一篇挺有意思的文章《写下记忆:理解、推导、扩展LSTM》(英文),从 RNN 的原理开始讲起,并讲述 RNN 为什么理论上很美好,实践起来要出问题,以及各种问题是怎么被解决的,从而诞生了 LSTM 。


这回应该不会忘记为什么 Basic LSTM 的模型是那样的了...






读例子程序: 1. ...


Udacity 上有一个 Google 技术人员提供的基于 Tensorflow 的深度学习课程,今天学到 Embeddings ,有点难理解,所以写个笔记记录下,以备日后回忆。


Udacity课程视频 这个课程在 Udacity 上的难度级别已经是 高 了。估计再下去就更少视频学习内容了。:~(






先搜索,再搜索。 把你找到的东西记下来。就算你在搜索的时候找不到其他有用的答案,把你搜到的跟你问题有关的链接放进来并说明为什么没有回答你的问题,这样能够让来回答的人更明白你问题的意思。


Recurrent Neural Network(RNN) Implementation

I heard about RNN for a long time, and have learned the concept several times, but until yesterday, I can’t implement any useful code to solve my own problem.

So I checked some tutorial. The...

from Tensorflow to Keras

I know 3 high level api for deep learning. They are Tensorflow.contrib.learn(SKFlow), TFLearn and Keras. All of them are great tools, but maybe I like Keras because of the easy style of code.


About sparse_column_with_hash_bucket

Yesterday, I saw tf.contrib.layers.sparse_column_with_hash_bucket in a tutorial. That’s a very useful function! I thought. I never met such a function in Keras or TFLearn.

Basically, the fun...

Thank you, GitHub and Jekyll

Thank you, GitHub and Jekyll.

Now I have my homepage again.

I’d like to share my thoughts and ideas and source codes here. Hope they helped.

btw, I am so happy to see Jekyll’s ...