# CSE517a Calendar (SP19)

## This is an inactive course webpage.

All course information and contents for SP2020 will be available on Canvas.

If you are wondering where we are going: here is a Roadmap!

#### Topic

15 Jan

17 Jan

Course Overview, Syllabus

Structural Risk Minimization

• lecture notes (.tex, .pdf)
• HERE is a folder with the images (remember to update this folder as we add new illustrations)
• FCML: Ch1, Linear Modelling
• ESL: 3.4.3, 10.6
22 Jan  Optimization

• GD (brief recap)
• Newton
• SGD
• momentum method
• FCML: Comments 4.1 & 2.6
24 Jan Estimating Probabilities form Data

• Coin Flipping
• MLE
• MAP
• FCML: 2.1-2.6, Random Variables and Probability
• FCML: 3.1-3.7, Coin Game
29 Jan  MLE and MAP for discriminative ML

• Linear Regression
• Ridge Regression
• Logistic Regression
• FCML: 2.8, MLE
• FCML: 3.8, 4.2-4.3, MAP
• FCML: 5.2.2, Logistic Regression
31 Jan   Squared Euclidean Distances

• Use-cases
• Matrix Equations
• Efficient Computation
5 Feb

7 Feb

Naive Bayes

• Generative ML
• Categorical features
• Multinomial features
• Continuous features (Gaussian NB)
• Missing features
• FCML: 5.2.1, Bayes Classifier and NB
• Ch 3 of Tom Mitchell’s ML book
12 Feb

14 Feb

Performance Evaluation

• Performance Measures
• Regression
• Classification
• Statistical Tests
• Cross-Validation
• Re-sampling
• lecture notes (.tex, .pdf, slides)
• FCML: 5.4, Performance
• ESL: 7.10, Cross-Validation
19 Feb  RBF Networks

• Kernel regression
• RBF network
• lecture notes (.tex, .pdf)
• LFD: eCh6.3-6.3.2, RBF Networks
21 Feb

26 Feb

Kernels

• Valid Kernels
• Kernel Construction
• Kernel Machines
• LFD
• 3.4, Non-linear Transformation
• 8.3, Kernel Trick & Kernel SVM
• FCML: 5.3.2, SVM and Kernel Methods
28 Feb  Recitation Session

• features for text data and text classification
• discussion of written hw1 and hw2 (selected problems)
For questions contact Zach and Trevor.

Mar 5

Mar 7

Gaussian Processes

• GPR via Weight-space View
• Definition: GP
• GPR via Function-space View
• Noise-free Observations
• Noisy Observations
• noisy Predictions
• GPR Algorithm
• Hyperparameter Learning
• lecture notes (.tex, .pdf)
• FCML: 8.1, Non-Parametric Models
• FCML: 8.2, GP Regression
• GPML: 2.2-3, Function-space View
• (*) GPML: 2.1, Weight-space View
• (*) GPML: 4.2, Covariance functions
• (*) GPML: 5, Hyperparameter Learning & Model selection
Mar 12

Mar 14

Spring Break
Mar 19 MIDTERM EXAM

• in-class
You can bring a cheat sheet with the following specifications:

• one US-letter sized page
• double-sided
• handwritten
Mar 21 Wrap-up: Kernel Methods and GPs

• GP Hyperparameter Training
• Multi-class Classification (for SVM, GPs)
• 1-vs-all
• Platt scaling
• 1-vs-1
• (*) Multi-class Logistic Regression
• Application: Traffic Prediction
• FCML: 8.4, Hyperparameter Optimisation
• GPML: 5.1, 5.3-4  Hyperparameter Learning & Model selection
Mar 26

Mar 28

Clustering

• Unsupervised Learning
• k-Means
• Kernel k-Means
• GMMs
• lecture notes (.tex, .pdf)
• FCML: 6., Intro
• FCML: 6.2, k-Means, Kernel k-Means
• FCML: 6.3.1-7, Mixture Models
Apr 2

Apr 4

Dimensionality Reduction

• PCA/SVD/MDS
• Non-linear Dimensionality Reduction
• Data Preprocessing
• Feature Engineering
• lecture notes (.tex, .pdf)
• FCML: 7.1-7.2, PCA
• LFD:
• eCH 9.1, Input Preprocessing
• eCH 9.2, PCA, SVD
• slides
Apr 9

Apr 11

Neural Networks – Basics

• Examples
• Feed-Forward NNs
• Back-Propagation
Apr 16

Apr 18

Learning (Deep) NNs

• Architecture
• Regularization
• Optimization
• Initialization
• Pre-training

Apr 23 Beyond FFNNs

• Autoencoders
• Recurrent NNs
• [optional] Convolutional NNs
Apr 25  Semi-Supervised Learning

• self-training & co-training
• GMMs
• graph-based SSL
• [optional] generative models
• [optional] S3VM
• (*) ESL: 17.1-17.3.1, Undirected Graphical Models
7 May Final EXAM

• 6-7pm (starts 6pm sharp!)
• be there by 5:50pm
• in Louderman 458
You can bring a cheat sheet with the following specifications:

• one US-letter sized page
• double-sided
• handwritten