- Author:
Yiqing Ma(Hongkong University of Science and Technology ) Deep Learning , Course
Course info
Instructor:
- qifeng chen (cqf@ust.hk)
TA:
- hyukryul yang(hyangbd@connect.ust.hk)
- nayeon lee (nayeon.lee@connect.ust.hk)
Score:
- Class participation: 10%
- In-class presentation: 15%
- Homework: 30%
- Final project: 45%
Website:
https://course.cse.ust.hk/comp6211d/Password_Only/index.html
- Week 1-2
Overview of Deep Learning: Architecture, Losses, and Optimization
- Week 3-4
Convolutional Neural Networks: Dilated Convolutions, ResNet, Perceptual losses
- Week 5
Deep 3D Vision: PointNet++, OctNet, Tangent convolutions
- Week 6-7
Graph Convolutional Networks for Graph Processing and Optimization
- Week 8-9
Sequential Modelling and Signal Processing: RNN, LSTM, TCN, and WaveNet
- Week 10-11:
Generative Models: GAN, Pix2pix, CycleGAN, CRN, VAE
- Week 12-13:
Final project presentation and project report due
Course
Depth: Repeated Composition
1st hidden layer - contour
Visible layer - pixel layer
Most node are represented in tensor => so called tensorflow
ML and AI:
Deeplearning example :MLPs
Representation learning example: Shallow autoencoders
Machine Learning example:Logisti regression
AI example:Knowledge bases
Historical trends: growing datasets It make deep learning works
Solving Object Recognition
Linear Algebra:
- Matrices
- Identity Matrix
- Systems of Equations Ax = b
- Matrix Inversion
- Norms
- Special Matrices and Vectors
- unit vector
- symmetri matric
- orthogonal matrix
- Eigendeomposition
- Eigenvalue
- effect of Eigenvalue
- Moore-penrose pseudoinverse
least square solution
- Computing the pseudoinverse
- Trace
- compute the value
- Learning linear algebra
- Probability Mass Function
- Variance and Covariance
- softmax->probability
- Bernouli Distribution
- Gaussian Distribution
- More Distribution
- Laplace