Deep Learning Literature Reading notes (4)

31. Convolution neural Network and its application in its vision (convolutional Networks and Applications in Vision) (English, conference papers, 2010, IEEE Search)

This paper introduces the principle and structure of CNN in detail, summarizes the application of convolutional neural network in many aspects, and gives the CNN unsupervised training improvement Program, and makes a lot of comparison and realization, the reference literature has authoritative.

32. Joint training of convolutional networks and non-convolutional networks (Joint training of convolutional and non-convoltional Neural Networks) (English, conference papers, 2014, IEEE Search)

This paper mainly introduces the application of convolutional neural network in speech recognition, and abstracts the depth structure of convolutional networks.

33, Sparse filter (Sparse Filtering) (English, conference papers, 2011, ei Search)

As a typical improvement of CNN training unsupervised, the sparse filtering algorithm, which belongs to the original literature of sparse filtering algorithm, and mentions the idea of combining CNN with SVM, is the necessary reference for the unsupervised improvement of CNN training.

34. Vehicle type identification using unsupervised convolutional neural networks (Vehicle type classification using unsupervised convolutional neural network) (English, conference papers, 2014, IEEE search)

This article uses convolutional neural networks for vehicle type identification, which is a new problem in old methods. At the same time, the unsupervised improvement of convolutional networks is explained very clearly, and the sparse filtering algorithm is used to break the traditional limitation of BP rules and is a typical application of sparse filtering method.

35. Visual target classification based on sparse convolutional neural Network (visual object classification by sparse convolutional Neural network) (English, conference papers, 2006, Google Scholar)

This article can be said to be the most primitive of sparse convolutional neural network, but not in detail, in which the convolutional neural network and the current study of the network in the structure also has a large difference, in writing the paper as the original document introduced some can.

36, based on the semi-supervised neural network vehicle type identification (Vehicle type classification using semi-supervised convolutional Neural network) (English, periodicals, 2012, EI search)

This article, which is a sparse representation and CNN, belongs to the improvement of the CNN convolution kernel, that is, the sparse Laplace filter learning method is used in the convolution kernel parameter learning.

37, convolutional Neural Network (Note on Convolutional Neural Network) (English, unknown, Google academic)

This is CNN's classic Basic article, the CNN Foundation gradient descent method is described in detail, and gives two key programming recommendations, and Deeplearning's MATLAB toolbox is matching, the only problem is that the article is not officially published, no retrieval.

38. Progress in the research of sparse representation method for target detection (Chinese, journal, 2015, net)

This paper introduces the application of sparse representation in target recognition, summarizes the framework of "feature extraction + classifier", introduces target feature learning, introduces target classifier, and explains the principle of sparse representation in detail, citing the full literature.

39. Convolutional Neural Network learning method based on small samples (learning convolutional neural network from few samples) (English, conference papers, 2013, ei Search)

This article belongs to the improvement of the method of initialization of convolutional neural network, that is, by using different initialization methods, the training of CNN is started in a relatively sparse position which is advantageous to the global optimal, so that the fast convergence of small sample training is realized.

40, PCA Pre-trained convolutional neural network (Chinese, periodicals, 2016, know-net)

This article is also an improvement in convolutional core initialization in convolutional neural networks, which is replaced by a mapping matrix consisting of the first few eigenvectors in the PCA transformation when initializing the convolution kernel, making the initialization more efficient. At the same time, it is mentioned that the next sampling is the probability maximization under the sampling method, which should be introduced in more detail in the reference literature. Finally, there is a picture in the text is very valuable reference.

Deep Learning Literature Reading notes (4)