深度学习

计算机科学系研究生选修课, 厦门大学, 2020

该课程为厦门大学计算机科学系研究生选修课,以基于深度学习平台TensorFlow和PyTorch的编程实践为主,算法理论为辅,使学生能够领悟深度学习的基本原理以及适用场景,并且对使用深度学习方法来解决问题具有一定的动手能力,为学生今后开展科研工作和业界求职打下基础。

教材

《Deep Learning》, Ian Goodfellow, Bengio Yoshua, and Courville Aaron, MIT press, 2016.

课程大纲

章节主要内容Notebook
Lecture 1 Introduction to Deep LearningDeep learning applications, impacts, researchers, history, and course description. 
Lecture 2 Basics of Machine LearningBasics of machine learning, linear models, neural networks, back-propagation, model selection, model evaluation.Lecture 2
Lecture 3 Regularization and OptimizationGeneralization, overfitting and underfitting, regularization, optimization for deep models, batch normalization, parameter initialization.Lecture 3
Lecture 4 Hardware and SoftwareDeep learning hardware, PyTorch, TensorFlow, Keras.Lecture 4
Lecture 5 Basics of Convolutional Neural NetworksConvolution, padding, stride, parameter sharing, pooling, common CNN patterns.Lecture 5
Lecture 6 CNN ArchitecturesAlexNet, VGG, GoogLeNet, ResNet, SENet, DenseNet. 
Lecture 7 Basics of Recurrent Neural NetworksRNN, Seq2seq, Attention models, LSTMLecture 7
Lecture 8 Language ModelWord2vec, ELMo, Transformer, BERTLecture 8
Lecture 9 Generative Adversarial NetworksGAN, DCGAN, CGAN, WGAN, SAGAN, pix2pix, CycleGAN, SRGAN 
Lecture 10 Deep Reinforcement LearningMarkov Decision Process, Q-Learning, Deep Q Network, Policy Gradient, Actor-Critic, DDPG.Lecture 10
Lecture 11 Deep Learning on GraphsDeepwalk, LINE, Node2vec, GCN, GraphSAGE, GAT 
Lecture 12 Self-Supervised LearningGeneration-Based Methods, Context-Based Methods, Free Semantic Label-Based Methods, Cross Modal-Based Methods, Contrastive Learning 

大作业展示

标题
A Lightweight Face Recognition Network with Bootstrapping
Adversarial Examples for Semantic Segmentation
Agricultural Product Price Prediction Based on Improved Temporal Convolutional Network
Application of Generative Adversarial Networks in Augmentation of Medical Image Datasets
Automatic Detection of Hard Exudates in Retinal Fundus Images
Construction and Prediction of Antimicrobial Peptide
Convert Screenshots of Breath of the Wild into Realistic Style using Cycle-Consistent Adversarial Networks
Convolutional Neural Networks for Identifying Human Behavior
Deep Learning Techniques for Eclipsing Binary Light Curves Classification
Explainable Channel Pruning for Accelerating Deep Convolution Neural Networks via Class-wise Regularized Training
Face Mask Detection Based on SSDv2 Network
Face Recognition Based on MTCNN and FaceNet
Facial Expression Recognition with Convolutional Neural Networks via a Data Augmentation Strategy
Generative Dog Images Using a Variety of GANs
Graph Neural Network for User-Item Recommendation
Identity Preserving Face Completion with Landmark based Generative Adversarial Network
Implementation of Text Detection and Recognition in Natural Scenes
Logit and Feature Dual-level Alignment for Visible-Infrared Person Re-Identification
Multi Crowd Mask Detection Method Based On SSD
Multi-Label Image Recognition in Anime Illustration with Graph Convolutional Networks
PointConv++ A Muti-Resolutional Network for Point Cloud Classification
pSp:A StyleGAN Encoder for Super Resolution
ResGCN A Method to Train Deep Graph Convolutional Network
Road Violation Detection
Semi-supervised Bert for Question Answering Matching
Sentiment Classification Based On BERT
Surgical Video Image Restoration Method based on Deep Learning
Takeaway Comments Sentiment Analysis Based on BERT
Text Classification Based on Bert

参考资料

本课程的课件参考了许多著名的深度学习课程,非常感谢这些课程的教授对课件进行无私的分享。

CS231n: Convolutional Neural Networks for Visual Recognition, Stanford University

CS224n: Natural Language Processing with Deep Learning, Stanford University

CS224w: Machine Learning with Graphs, Stanford University

CMSC 35246: Deep Learning, University of Chicago

Introduction to Reinforcement Learning with David Silver, DeepMind

MGMTMSA-434: Advanced Workshop on Machine Learning, UCLA

期末合影

IMG_7417