【论文学习】深度学习“里程碑”论文

https://github.com/floodsung/Deep-Learning-Papers-Reading-Roadmap

1.1 Survey

[1] LeCun, Yann, Yoshua Bengio, and Geoffrey Hinton. “Deep learning.” Nature 521.7553 (2015): 436-444. [pdf] (Three Giants’ Survey) :star::star::star::star::star:

机器学习被广泛用于图像、语音、推荐。传统机器学习需要小心的构造特征。
深度学习是一种 Represention learning，可以自动构建特征。它在一层构建特征，然后输出给更抽象的下一层。
以图像为例，第1层检测不同角度和位置的 edge 是否存在，而第2层能检测图案，第3层可能把图案组合起来，如此等等。
深度学习在解决以前的一些难题上是重大突破，它可以很好的发现高维复杂的特征，所以广泛用于科学、商业、政府。在 image recognition、speech recognition 等方面都打破了记录，还在药品分子建模、脑科学、DNA开合预测方便击败其它机器学习模型。在NLP领域，用于 topic classification, sentiment analysis, question answering，translation

有监督学习：设定一个优化目标（得分或距离），然后用 SGD 去优化参数。

现代的神经元基本都用 ReLU，它训练更快，历史上用 sigmoid 更多。

以前人们认为局部最优是一个问题，近期人们发现更大的问题其实是 saddle points

NLP领域

• 2006年 加拿大CIFAR 团队引入了无监督过程，整个系统可以用 backpropagation 做 fine-tune
• 2009年，在GPU上训练 语音识别任务，加快了10～20倍，打破了记录

CV领域。filter layer 可以用来提取特征。pooling layer 也需要，因为 filter layer 提取的特征可能差别很大，

词向量之前，普遍的做法是 n-gram

RNNs 是很强大的动态系统，但是训练困难，因为面临梯度爆炸和梯度消失的问题

翻译模型：先用一个 encoder 把整个句子意义提取成一个向量，然后用 decoder 把这个向量翻译成另一种语言

1.2 Deep Belief Network(DBN)(Milestone of Deep Learning Eve)

[2] Hinton, Geoffrey E., Simon Osindero, and Yee-Whye Teh. “A fast learning algorithm for deep belief nets.” Neural computation 18.7 (2006): 1527-1554. [pdf](Deep Learning Eve) :star::star::star:

[3] Hinton, Geoffrey E., and Ruslan R. Salakhutdinov. “Reducing the dimensionality of data with neural networks.” Science 313.5786 (2006): 504-507. [pdf] (Milestone, Show the promise of deep learning) :star::star::star:

借助多层神经网络，高维数据可以转为低维数据。
在这个“autoencode” 中，可以用 gradient descent 来优化，前提是 initial weights 已经接近最优解了。

本文介绍了一种初始化 weights 的方案，使得 autoencoder 表现效果比 PCA 好很多

1.3 ImageNet Evolution（Deep Learning broke out from here）

[4] Krizhevsky, Alex, Ilya Sutskever, and Geoffrey E. Hinton. “Imagenet classification with deep convolutional neural networks.” Advances in neural information processing systems. 2012. [pdf] (AlexNet, Deep Learning Breakthrough) :star::star::star::star::star:

[5] Simonyan, Karen, and Andrew Zisserman. “Very deep convolutional networks for large-scale image recognition.” arXiv preprint arXiv:1409.1556 (2014). [pdf] (VGGNet,Neural Networks become very deep!) :star::star::star:

[6] Szegedy, Christian, et al. “Going deeper with convolutions.” Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition. 2015. [pdf] (GoogLeNet) :star::star::star:

[7] He, Kaiming, et al. “Deep residual learning for image recognition.” arXiv preprint arXiv:1512.03385 (2015). [pdf] (ResNet,Very very deep networks, CVPR best paper) :star::star::star::star::star:

1.4 Speech Recognition Evolution

[8] Hinton, Geoffrey, et al. “Deep neural networks for acoustic modeling in speech recognition: The shared views of four research groups.” IEEE Signal Processing Magazine 29.6 (2012): 82-97. [pdf] (Breakthrough in speech recognition):star::star::star::star:

[9] Graves, Alex, Abdel-rahman Mohamed, and Geoffrey Hinton. “Speech recognition with deep recurrent neural networks.” 2013 IEEE international conference on acoustics, speech and signal processing. IEEE, 2013. [pdf] (RNN):star::star::star:

[10] Graves, Alex, and Navdeep Jaitly. “Towards End-To-End Speech Recognition with Recurrent Neural Networks.” ICML. Vol. 14. 2014. [pdf]:star::star::star:

[11] Sak, Haşim, et al. “Fast and accurate recurrent neural network acoustic models for speech recognition.” arXiv preprint arXiv:1507.06947 (2015). [pdf] (Google Speech Recognition System) :star::star::star:

[12] Amodei, Dario, et al. “Deep speech 2: End-to-end speech recognition in english and mandarin.” arXiv preprint arXiv:1512.02595 (2015). [pdf] (Baidu Speech Recognition System) :star::star::star::star:

[13] W. Xiong, J. Droppo, X. Huang, F. Seide, M. Seltzer, A. Stolcke, D. Yu, G. Zweig “Achieving Human Parity in Conversational Speech Recognition.” arXiv preprint arXiv:1610.05256 (2016). [pdf] (State-of-the-art in speech recognition, Microsoft) :star::star::star::star: