Convolutional Neural Network Study Notes and Experience (1) Preface

The blogger is a one-year undergraduate student. This time last year, I had the privilege of working on an experimental project on character recognition, a rare opportunity for an undergraduate who was planning to work on data mining and machine learning. When I first came into contact with this project, my colleagues and I were at a loss and didn’t know where to focus. In addition, I made a simple character recognition system at the undergraduate level. Too much attention is paid to the recognition link, but a lot of energy and time are spent on image preprocessing, error correction and other links.

After a period of exploration, our focus returned to identification. About 4 months ago, deep learning caught our attention. At that time, the concept of deep learning was too hot. Just a few months before we settled on the model, AlphaGO defeated several Go masters, especially Li Sedol. When the results were announced, the world was shocked. Since deep learning has such magic power, it seems that in the future research on machine learning, deep learning cannot be avoided. Therefore, we decided to introduce deep learning into our project.

From knowing nothing to being able to make some subtle improvements to other people's models, I have stepped on a lot of pits in the process, and of course I have a lot of insights. Therefore, I plan to write down what I have learned and experience to share with you. I mentioned that I am an undergraduate student, and I have no teachers in related fields to help me. Therefore, when I look at the problem, I may not be as comprehensive as those great gods, and I may not be able to analyze the problem. , If the blog post is not comprehensive, incomprehensible, or wrong, please do not hesitate to let me know.

Finally, let me talk about my experimental environment:
* CPU: Intel(R) Core(TM) i5-4460 CPU @ 3.20GHz
* GPU: None
* Memory: 32G
* Operating system: Ubuntu14.04
* Deep learning framework: caffe docker version