Deep Learning tips-deep learning

Entry route
1, first of all on their own computer to install an open source framework, like TensorFlow, Caffe such, play this framework, the framework to use
2, and then run some basic network, from the
3, if there are conditions, the entire GPU computer, GPU run a lot faster, compared to the CPU

To be more specific, I think you can follow these steps to learn it:
First phase:
1, realize and train only one layer of Softmax regression model for handwritten digital image classification;
2, the implementation and training of three full-link layer model for handwritten digital image classification;
3), the implementation and training of three rolls of grass-roots + pool layer model for handwritten digital image classification;
Stage Objective: To understand some machine learning, in-depth learning concepts, find an open source tool and try to train some simple network, try to join some common trick to debug the network.

Second phase:
Training some classic CNN classification network, familiar with some common data sets, familiar with the development of the CNN classification network, some trick replacement:
1), lenet,1986 years
2), alexnet,2012 years
3), googlenet,2014 years
4), vgg,2014 years
5, ResNet, also some people call the residual network, 2015
This network is a classic of deep learning in the field of images, implemented on PCs, trained on them, and you look at their network structure.
Stage Objective: They can be used independently in practice, and they may also be used in combination with traditional image processing techniques. In a word: you're already in.