📗 -> 05/05/25: ECS189G-L16

Midterm Overview

🎤 Vocab

❗ Unit and Larger Context

Midterm

Double side A4 cheat sheet

Coverage:

Intro
ML basics
Optimization
DL basics
Auto-encoder (VAE optional)
CNN
Computer vision
GAN

Format

TF questions with short explanation
Short answer questions
Calculation Questions

✒️ -> Scratch Notes

Review

Section 1: Intro

Where is it applied?
Market forecast
Industry and academia achievements

Section 2: ML Basics

Machine learning overview
Supervised: Classification

Train/test split
SVM
Supervised: Regression
Linear regression, Ridge, Lasso
Unsupervised: Clustering
K-means
Evaluation Metrics

Section 4: DL Optimization

Data perspective:

Inputs: full batch, instances, mini-batch
Output (real value, probability, etc.): Decide loss function
Design model:
Initialize variables to be learned
- Abstracted by pytorch
  Decide optimizer
GD, Momentum, Adagrad, Adam, vs …
Specify optimizer parameters
- Learning rate
- Other parameters

Section 5: DL Basics

Background:

What is it?
Why do we need?
Brief history
What makes it successful
Technical Details:
Biological Neuron vs Articial Neuron
Perceptron and its weakness
- Exposed by Minsky, XOR?
Multilayer perceptron and applications
How to train MLP
ERROR BACKPROP ALGORITHM

Section 6: Auto-encoder

Auto-encoder

What is it?
What is the architecture
How to train the model?
- e.g. loss function definitions, …
What can autoencoders do and what are their advantages
What are the potential applications of auto-encoders
VAE (optional)
More like a probabilistic graphical model
Why called auto-encoder
What is VAE model architecture and learning process?
(also optional) VAE learning with gradient descent via re-parameterization

Section 7: CNN

Conventional Image Processing

Weakness of MLP for images
Image processing operations with convolution operator
What is convolution?
How to calculate convolution on images?
Several key parameters (kernel size, stride, padding, channel, etc…)
CNN:
What is CNN architecture
Pooling layer?
CNN and its variant
how to make deeper?
what is residual learning? What is resnet?

Section 8: Computer Vision

What is CV?

What about
representation
difficulties
Problems (optional)
Detection, localization, segmentation
image classification, style transfer, colorization, super-resolution, reconstruction, synthesis
What are proposed solutions (optional)
Examples proposed based on CNN

Section 9: GAN

Discriminative models vs generative models

Examples of D models
Can we jsut use auto-encoders as the G?
Why do we need to study G models?
Generative Adversarial Model (GAN)
Game between G & D
How to train model iteratively
What can GAN do?
Examples

Q&A:

Can an SVM be a discriminator?

Yes it can
Tries to distinguish different instances/classes
How much math (in lasso, SVM, etc.)?
no equation writing, but…
come for deep learning, ML need to know. not incorporated here though.
Similar to HW?
More comprehensive than that
Calculations we need to know?
Yes one, but he can’t say
The history of DL?
History is motivation for material
He hopes we know it, but not tested

🧪 -> Refresh the Info

Did you generally find the overall content understandable or compelling or relevant or not, and why, or which aspects of the reading were most novel or challenging for you and which aspects were most familiar or straightforward?)

Did a specific aspect of the reading raise questions for you or relate to other ideas and findings you’ve encountered, or are there other related issues you wish had been covered?)

Vault

Explorer

ECS189G-L16