1、terminology what is the function of a fully connected layer?
2、terminology what is a recurrent neural network?
3、deep neural networks. see the figure above: describe what the numbers (224, 11, 96b etc.) and the words mean in the enclosed area (the first 3 layers).
4、deep neural networks. give 4 examples for how to perform image augmentation.
5、mathematics of deep learning. give the formula for the convolution of an digital image l(i,j) with filter f(i,j).
6、mathematics of deep learning. what is an affine transformation? also give the formula.
7、face recognition. why are histograms of gradients (hog) working so well in face detection?
8、human vision. what is the function of the filters in the first visual layer, v1 (also called the primary visual cortex)? if possible, make a drawing of these filters.
第一章 深度学习入门和基本原理
chapter 1 exercise
1、deep learning is a class of ______ algorithms. a. rote learning b. learning by analogy c. learning from induction d. machine learning
2、deep learning algorithms type(s) are ______. a. supervised b. unsupervised c. supervised and unsupervised d. neither
3、what do filters do? a. convolution b. pooling c. max-pooling d. deconvolution
4、which is not deeping learning frameworks? a. caffe b. tensorflow c. seqan d. torch
5、‘deep learning’ means using a neural network with____ between input and output. a. several layers of nodes b. some nodes of layers c. neural cells d. cnn
第二章 卷积神经网络
chapter 2 exercise
1、which is not the activation function？ a. relu b. elu c. tan(x) d. sigmoid
2、instead of just looking at overall accuracy, we calculate _____ metrics. a. revising b. precision and recall c. forecasting d. prediction
3、deep nets are made by stacking learned ___ layers and simple pointwise ___ layers a. linear, non-linear b. non-linear, linear c. linear, linear d. non-linear, non-linear
4、output is a function of every input, or the input and output are ____ a. fully connected b. incomplete connected c. one-to-one link d. natural link
5、in machine learning, having more ___ is almost always more important than having better algorithms. a. data b. powerful gpu c. filters d. kernels
第三章 基于cnn的人脸识别
chapter 3 exercise
1、which situation is not convenient to utilize alexnet? a. classification of tumor b. classification of dogs & cats c. natural language understanding d. vision of self-driving taxi
2、using back propagation, the is? where a=2, b=1,c=3,d=2,e=6. a. 1 b. 2 c. 3 d. 4
3、which is not the platform for deep learning? a. pytorch b. tensorflow c. python d. maxnet
4、what's the name of this picture? a. gabor kernel b. gauss kernel c. relu function d. sigmoid function
5、what's the name of this picture? a. gabor kernel b. gauss kernel c. relu function d. sigmoid function