2147334 Machine Learning or Deep Learning

1. Zoom link: https://chula.zoom.us/j/96641838331?pwd=ZVBuMGJTVnJlbnNLdGRiVm02QWRDZz09

2. Shared Drive: https://drive.google.com/drive/folders/136FIuxa4JNiLZu4vAYd4eQZiqBCZxLKO?usp=sharing

3. Discord Channel: https://discord.gg/7vE3NCJy

4. Schedule:

Topics

9-Aug-21

Course Introduction (Link to Pre Course Suvey, Link to Covid19)

16-Aug-21

Hypothesis Representation (VDO Playlist)

1. Introduction to Machine Learning (16:07 mins)

Objectives: Know the brief concept of Supervised learning, Unsupervised learning, and Reinforcement Learning

2. Hypothesis Representation (15:32 mins)

Objective: See some examples of hypothesis presentations in 2D space

3. Introduction to Sigmoid Function (7:20 mins)

Objective: Know the concept of Sigmoid Function

4. Behavior of Sigmoid Function (15:05 mins)

Objective: Introduction to a parameter tuning

5. Curve Fitting, Trend Prediction using Sigmoid Function (Python Coding) (23:24 mins)

Objective: Can see and example of parameter tuning using sigmoid function

6. Hypothesis Space (16:09 mins)

Objective: Know the size of hypothesis space

7. Live Session (Start from 15:00):

Objective: Another example of hypothesis representation of COVID-19 data (Linear Regression)

Colab link: https://colab.research.google.com/drive/15DMZQUJPG1K-GKXPcUzY4af_QXlaWdCc?usp=sharing

Recorded Zoom: https://youtu.be/2gmyyyxWuok

23-Aug-21

Version Space (VDO Playlist, Recorded Zoom)

1. Intro to Version Space (6:14 mins)

2. Data and Hypothesis Representaion (4:41 mins)

3. Explanation of Hypothesis Representation (6:13 mins)

4. Size of Hypothesis Space (8:57 mins)

5. General and Specific Concept from the text book slides (6:03 mins)

6. Find-S Algorithm (6:12 mins)

7. Demonstration of Find-S (6: 31 mins)

8. Search Lattic of Find-S (8:43 mins)

9. Bias of Find-S (6:06 mins)

10. Intro to Version Space (the Candidate Elimination Algorithm( 8:56 mins)

11. An Example of the Version Space Algorithm (15:57 mins)

12. Live Session Starts from 15:30

30-Aug-21

Decision Tree Learning (VDO Playlist)

1. Theory

   1. Introduction to Decision Tree (10:54)

   2. Do You Have a Measurement (6:22)

   3. Introduction to Entropy (14:25)

   4. Finding Entropy (9:55)

   5. Build a Full Tree (8:05)

2. Writing Program (17:19, 7:50, 26:26, 14:37)

3. Q&A Live Session Starts from 15:30

6-Sep-21

Hypotheses Evaluation (VDO Playlist)

1. Know components in confusion matrix

2. Know meaning of precision and recall

3. Know what are FPR, FNR, TPR, TNR

4. Can write a program which can calculate the confusion matrix of a testing

5. Can write a program to compare two learning algorithms

6. Q&A Live Session Starts from 15:30 (Recorded Zoom)

   1. Please evaluate DTL and KNN on IRIS data

      1. Construct a confusion matrix for both DTL and KNN using train_test_split()

      2. Using 5-fold Cross Validation to compare both algorithms. Please find the significant level also.

13-Sep-21

Neural Networks (1) (VDO Playlist)

1. Know the structure of Neuron

2. Know that a Neuron is equal to a linear in 2D space

3. Can find a weight vector that classify the OR table

4. Know how to train a perceptron using a spreadsheet

5. Know a sigmoid unit

6. Know how to train a sigmoid unit

20-Sep-21

Neural Networks (2) (VDO Playlist)

1. Can proof how to train a TLN

2. Can proof how to train a Linear Node

3. Can apply training concept to other kinds of nodes

27-Sep-21

Midterm Exam Week (No Class)

4-Oct-21

Neural Networks (3) (VDO Playlist)

1. Can proof how to train a Sigmoid Node

2. Know concept of Gaussian Node

3. Can proof how to train a Gaussian Node

4. Live Session starts from 15:45 for only Q&A (No new content or examples will be added)

11-Oct-21

Quiz#1 and Neural Network (4)

Quiz#1 (less than one hour)

1. Hypotheses Representation (Create your own hypothesis representation on your selected data. Should not be ever used in our class)

2. Parameter Tuning Using curve_fit function

3. Show the performance of your hypothesis on the unseen data

4. Submit your ipynb file before 14:00, 11 October 2021, via a google form => HERE.

5. A reference is in 16 Aug 2021 class.

Watch this VDO Playlist

1. Can design a MLP

2. Can train a BNN

18-Oct-21

HW#1

1. Colab to Decision Tree: https://colab.research.google.com/drive/1ulvs_AsKmU_PIPwnVAxx7Sy8PaOVPfyR?usp=sharing

2. TLN and Sigmoid: https://colab.research.google.com/drive/1DUE9jynrou13u9_0HvvGQ6XftWVDdNYy?usp=sharing

3. Submit a photo via this form before 23:59, 18 Oct 2021:

https://forms.gle/pbsngcLAmCojNj3X7

4. PDF written in live session: https://drive.google.com/file/d/1CPxkMqXrrDA3yuXfLIFQdTPnit_ncn73/view?usp=sharing

25-Oct-21

Quiz#2 Version Space and Decision Tree Learning (Please use 23 and 30 Aug VDOs as the references). This time, the quiz will be 3-4 questions relating to Version Space and Decision Tree. The quiz will be done in the class.

The quiz will start at 13:15. Here is the link.

Naive Bayes (VDO Playlist)

1. Naive Bayes Classifier: Bayes Theorem, A Maximum Posterior Hypothesis

Learn Bayes Theorem and Maximum Posterior Hypothesis. When you know some fact, find the hypothesis matches that fact.

2. Naive Bayes: An ATK Example

Learn an example using the "Real World Maximum Posterior Hypothesis". However, you have to think about the case of P(covid19) of the population and how does this probability change if you decide to take the test. If you decide to take the test, it means that your chance to be the covid19 class is much highter than the P(covid19).

3. Naive Bayes Classifier

Learn the concept of Naive Bayes Classifier. Try to understand the example shown in the VDO.

4. Text Classification Using Naive Bayes

Learn the steps of applying the Naive Bayes Classifier to classify text.

5. Source Code Explanation

You should be able to apply the source code to attack your real problem.

1-Nov-21

Clustering (SOM, K-Means)

The VDO Playlist is here.

Colab to HW. (Deadline 23:59, 1 Nov 2021)

1. Know the concept of K-Means, SOM.

2. Know how can we use the KMeans library from sklearn.

3. Understand the SOM code.

4. Can apply both algorithms to a new problem

8-Nov-21

Quiz #3 Neural Networks

You are supposed to show a network structure, given a set of positive and negative examples. Train a simple perceptron (TLN or linear). You can use the VDO on 13, 20 September 2021.

This quiz will be 30 minutes (submit before 13:45). Here is the link to the questions, https://docs.google.com/document/d/1WJZBqyrWHh9npThGBH08aPXMUB6LNw9VWjkTWFkiNeQ/edit?usp=sharing .

Deep Learning (CNN) (VDO Playlist)

15-Nov-21

LSTM (VDO Playlist)

Homework (Link to submit) (Deadline 23:59, 15 Nov 2021)

1. Using LSTM to predict some regression data with only 1 dimension (Covid19, Weather, etc) or

2. Using LSTM to learn text classification

22-Nov-21

Quiz#4 Naive Bayes Classifier and Project Presentation. You can find the Quiz here.

22 November - 10 December

Project Presentation: Put Everything Together.

1. Apply your knowledge from our class to attack the problem that you are interested

   1. One technique (Curve Fitting, Decision Tree, Neural Networks, CNN, LSTM, Clustering, Naive Bayes, etc.) 20 Marks

   2. Two or more techniques (Comparing or Combining 2 or more techniques) 25 Marks

2. The problem "must" be your own data (or get raw data from the internet)

   1. Your own data or Get raw data from the internet or Covid19 with some data processing 25 Marks

   2. Data (excel, json, csv) from the internet 20 Marks

   3. Data from kaggle.com 15 Marks

3. Analyze results from your technique

   1. Without results discussion (20 Marks)

   2. Can show your understanding of the obtained results (25 Marks)

4. Fine-tune your parameters

   1. Do only one experiment (No parameter tuning) (15 Marks)

   2. Can adjust the parameters to get better results (20 Marks)

   3. Add other techniques to improve the result (25 Marks)

5. You have to present individually (Please reserve your time slot, here)

29-Nov-21

Final Exam Week (No Class)

5. Grading

   1. Midterm (online) 30% (7.5+7.5+7.5+7.5)

   2. Final Project 40%

   3. Homework 30%

6. Textbook:

   1. Tom Mitchell, "Machine Learning", McGraw Hill, 1997.