Learning to be Giant.

Kaffe: 从零开始写一个神经网络(一)

|

神经网络(Neural Network)从被提出到现在已经三十多年,经历了兴盛与衰败,如今随着计算能力和数据量的增大又重返大家的视野。三十年间,无数的模型被提出:Convolutional Neural Network, Deep Belief Network, Restricted Boltzmann Machine, LSTM等等。模型种类纷繁众多,但最基本的构建模块却从来没有变过,包括Back Propagation和Stochastic Gradient Descent。本系列目录见这里

Kaffe: 从零开始写一个神经网络(零)

|

当我开始写这篇日志的时候无意间发现,距离我上一篇日志发布已经一整年了。那时的我还在暑期实习,趁工作间隙把我的博客重新整理了整理,开源了代码,如今也已经有一百多个star了。这一年发生了很多事情,我从CMU毕业了,成为了一名全职的机器学习工程师。深度学习这个领域也更加的火热。作为一个机器学习工程师,我想我有必要自己把深度学习最基础的问题搞懂。

Website Upgrade

|

During the past week, I made some upgrade to this website:

  1. Website currently is Jekyll 3 ready. During hte past several months, the website cannot be updated since Github Pages migrated from Jekyll 2 to Jekyll 3, and the website had some configurations that are not compatible with Jekyll 3.
  2. Stylesheets are refactored to Sass. The original Lanyon template used pure CSS. In this upgrade, I changed it to scss. Therefore, the current template is easier to maintain and customize. All the configuration that are relevant to style is in _config.scss.
  3. Support multilevel menu. The sidebar currently support multilevel menu (however only 2 levels supported).
  4. Better _config.yml. Now _config.yml contains more configuration options. People can get what they want without dive into the code.
  5. Many more…

The source code is https://github.com/codinfox/codinfox-lanyon.

How to compile Caffe

|

Compiling caffe is really a pain in the ass. This may be a slightly better (maybe not) version of the Caffe installation instruction. Please check http://caffe.berkeleyvision.org/installation.html for more information. This tutorial is tested on a workstation with Titan Black and Tesla K40. Ubuntu 14.04 Linux is assumed.

First of all, download Caffe source code from Github:

git clone https://github.com/BVLC/caffe.git

Center-Surround Mechanism for Edge Detection

|

在研究计算视觉模型的时候,Center-Surround机制是不得不提的,基本上是所有理论的前提。感性理解,Center-Surround机制就是通过对于receptive field当中不同位置对光的不同反应帮助生物视觉系统识别边缘信息。很早之前在做视觉注意(Visual Attention)的时候经常遇到Center-Surround,但是从来没有认真弄懂,只是默认它就等同于LoG(Laplacian of Gaussian)。如今又接触,终于了解清楚,于是记录下来。