08408005 Deep Learning: Algorithm and Application

Instructor:

Teaching Assistant:

Location: Room 206, Teaching Building 3, Peking University

Time: Monday 10:10am - 12:00am (weekly), Tuesday 10:10am - 12:00am (biweekly)

Office hours: Drop an email or wechat message to the instructor or TAs for appointing a course-related face-to-face QA.

Schedule of Lectures

Date	Topics
	Introduction Course logistics A short tour to deep learning: illustrative applications, brief history etc.
	Machine Learning Basics Machine learning paradigms Supervised learning: linear case Supervised learning: non-linear case Gradient Descent Stochastic Gradient Descent Unsupervised learning
	Deep Learning Basics Neural layers: convolution, pooling, FC, activation, loss functions Gradient back-propagation Auto-differentiation Regularization
	Backbone - I Conventional Image Classification AlexNet, VGGNet, GoogleNet, ResNet, DenseNet Squeeze-and-Excitation, GroupConv, Depthwise / Depthwise Separable Convolution, ShuffleNet
	Backbone - II Classic Object Detection Z-F Net RCNN family: RCNN, Fast RCNN, Faster RCNN YOLO, SSD RetinaNet, FPN etc
	Backbone - III Fully convolutional networks (FCN) Encoder-Decoder (SegNet), DeepLab Contextual inference with graphical models (CNN-as-RNN) Mask R-CNN Generalization to other pixel-to-pixel tasks Resolution matters: Simple Baseline & HRNet
	Deep Models for Sequential Data RNN LSTM Transformer MLP-Mixer Mamba
	Foundation Models
	Generative Models
	Reinforcement Learning
	Holiday - no class
	Learning on Graphs / Adversarial and Backdoor Learning
	DL for Science / Optimization
	Course Project Presentation - I
	Course Project Presentation - II
	Course Project Presentation - III

Textbook:

References:

Christopher Bishop, Pattern Recognition and Machine Learning, Springer, 2006
Shai Ben-David and Shai Shalev-Shwartz, Understanding Machine Learning: From Theory to Algorithms, 2014
Richard Sutton, Reinforcement Learning, 2nd edition
Zhi-hua Zhou, Machine Learning (in Chinese), 2016

Course Work

Final Grade	Grading will be based on homeworks (35%), survey (25%) and a course project (40%). The end-of-term grade is curved. Your overall grade will depend on your performance relative to your classmates.
Survey & Project	The topic of survey paper should be DL-related. A Course Project is mandatory for all students. The only requirement for the topic is being deep learning related.