CH1 Using neural nets to recognize handwritten digits
1. RNN的wiki:recurrent neural networks
2. Cauchy-Schwarz不等式的wiki:Cauchy-Schwarz inequality
CH2 How the backpropagation algorithm works
1. 提出backpropagation算法的paper:Learning representations by back-propagating errors
CH3 Improving the way neural networks learn
1. Cross entropy的wiki:brief summary
2. 一个哲学问题的wiki:problem of induction
3. paper:Gradient-Based Learning Applied to Document Recognition
4. paper:ImageNet Classification with Deep Convolutional Neural Networks
5. paper:Improving neural networks by preventing co-adaptation of feature detectors
6. paper:Best Practices for Convolutional Neural Networks Applied to Visual Document Analysis
7. paper:Scaling to very very large corpora for natural language disambiguation
8. paper:Practical Recommendations for Gradient-Based Training of Deep Architectures
9. paper:Deep, Big, Simple Neural Nets Excel on Handwritten Digit Recognition
10. paper:Random search for hyper-parameter optimization
11. paper:Practical Bayesian optimization of machine learning algorithms
12. paper:Practical recommendations for gradient-based training of deep architectures
13. paper:Efficient BackProp
14. paper:Neural Networks: Tricks of the Trade
15. paper:On the importance of initialization and momentum in deep learning
16. wiki:L-BFGS
17. paper: Understanding the difficulty of training deep feedforward networks
18. paper:What is the Best Multi-Stage Architecture for Object Recognition?
19. paper:Deep Sparse Rectifier Neural Networks
20. paper:Rectified Linear Units Improve Restricted Boltzmann Machines
CH4 A visual proof that neural nets can compute any function
1. paper:Approximation by superpositions of a sigmoidal function
2. paper:Multilayer feedforward networks are universal approximators
CH5 Why are deep neural networks hard to train?
1. paper:On the correlation of parity and small-depth circuits
2. paper:On the number of response regions of deep feed forward networks with piece-wise linear activations
3. paper: Learning deep architectures for AI
4. paper:Gradient flow in recurrent nets: the difficulty of learning long-term dependencies
5. paper:Untersuchungen zu dynamischen neuronalen Netzen
6. paper:Understanding the difficulty of training deep feedforward neural networks
CH6 Deep learning
1. paper:Visualizing and Understanding Convolutional Networks
2. DL相关网站: LeNet-5
3. Theano:Theano: A CPU and GPU Math Expression Compiler in Python
4. Classification datasets results,含有大量数据集和paper:informative summary page
5. paper:Building high-level features using large scale unsupervised learning
6. paper:ImageNet: a large-scale hierarchical image database
7. 图像数据集网站:ImageNet
8. paper:ImageNet classification with deep convolutional neural networks
9. paper:Theano-based large-scale visual recognition with multiple GPUs
10. paper:Going deeper with convolutions
11. paper:ImageNet large scale visual recognition challenge
12. paper:Multi-digit Number Recognition from Street View Imagery using Deep Convolutional Neural Networks
13. paper:Intriguing properties of neural networks
14. paper:Deep Neural Networks are Easily Fooled: High Confidence Predictions for Unrecognizable Images
15. paper:A fast learning algorithm for deep belief nets
16. paper:Reducing the dimensionality of data with neural networks
17. paper:RNN A 2014 paper
18. paper:RNN A second paper, also from 2014
19. LSTMs:Hochreiter and Schmidhuber in 1997
20. paper:to recognize shapes, first learn to generate images.
21. DBNs:this overview
22. DBNs:this article
23. paper:natural language processing
24. paper:also this informative review paper
25. paper: music informatics
26. paper:play video games well this followup
27. paper:annotate photos and videos