●子模組一：卷積神經網路 (Convolutional Neural Networks) 課程大綱： - Introduction
- Convolution Operation
- Pooling
- Variants of Basic Convolution Function
- Structured Output
- Data Types
- Efficient Convolution Algorithms
- Random or Unsupervised Features
- Neuroscientific Basis for Convolutional Networks
實驗： Convolutional Neural Network 實驗內容說明： Use CIFAR-10 Dataset and build a convolutional neural network to do multi-class classification In this problem, you will construct a Convolutional Neural Network (CNN) for image recognition using CIFAR-10 dataset. The CIFAR-10 dataset consists of 60,000 color images in 10 classes, with 6,000 images per class. There are 50,000 training images and 10,000 test images (a) Implement a CNN for image recognition using CIFAR-10. Analyze the effect of different settings including stride size and filter size. You should show the learning curve of training set and the final test error rate. (b) Show some of feature maps in hidden layers ●子模組二：自編碼器 (Autoencoders) 課程大綱： - Introduction
- Undercomplete Autoencoders
- Representational Power, Layer Size and Depth
- Stochastic Encoders and Decoders
- Denosing Autoencoders
- Learning Manifolds with Autoencoders
- Predictive Sparse Decomposition
- Applications of Autoencoders
實驗： Autoencoder 實驗內容說明： Use MNIST-M Dataset and build an autoencoder to observe the learning curve and the generated samples (a) Please build an autoencoder to reconstruct mnist-m dataset, and show the reconstruction loss during training stage (b) Please refer to the “adversarial autoencoder”(AAE), implement an AAE and use t-SNE dimension reduction method to plot the encoding of training data |