Sun, Dec 31, 2017
Read in 2 minutes
Lately I have been started learning on what and how deep learning works.
Deep learning is one of the current trends in IT technologies. Google have exponentially increased the number of product they are developed using deep learning which is kind of interesting to me. I tried to develop my own deep learning model before and stuck on how and what actually happens in it since deep learning or neural network is a black box model.
I slowly start learning from the top by cloning someone else codes and understanding what happening and understand the reason behind it by sometime reading their blog post or from research paper but still stuck at some point. I stop trying someone else code and try to learn basic math and statistics by reading and watching online materials such as “The Cartoon Guide to Statistics”, “Manga Guide to Statistic”, “Manga Guide to Linear Algebra” and “Manga Guide to Regression Analysis” for reading and “Coursera: Machine Learning”,“Udacity:Deep Learning” and “Udacity: Statistics” for watching videos. After completing and half complete(need some time to get back to it :p )
I can have confidence and understand what is the underlying concept behind deep learning. The reason why I started interested in Deep Learning was because of Style Transfer which is by combining two different image which is imageA(content image) and imageB(style image) into one different image that has imageA content and imageB style. At first, I read a blog post explaining about style transfer and understand the algorithm behind it but I stuck at how they even think about it.Later I found a blog post about feature visualization which is exploring all the different layer in CNN.
Every layer in CNN has their own importance in identifying different kind of image. The very first layer in CNN consist of small line or edge which can be used as edge detection and the layer will be combine in more “near object look”.