CSE 445: Machine Learning

Introduction

Instructor: Intisar Tahmid Naheen, North South University

 Resources
▪ Slides provided in course should be enough – but there is a plethora of
fantastic resources available, so use them!
▪ Recommended Books:
▪ Hands-On Machine Learning with Scikit-Learn, Keras, and Tensorflow by Aurelien
Geron (will be followed extensively in the course with code examples from
https://github.com/ageron/handson-ml )
▪ Pattern Recognition and Machine Learning by Christopher Bishop (excellent
resource for mathematical foundations)
▪ Elements of Statistical Learning by Jerome Friedman et al (good reference)
▪ Additional Material:
▪ Andrew Ng’s course on Machine Learning available on Coursera
▪ CS 189, Berkeley
▪ CS 229, Stanford
 Helpful Prerequisites
▪ MAT361- Probability & Statistics
▪ Probability distribution, Random Variable, Conditional Probability, Variance (some
of the important concepts to recall to name a few)
▪ MAT125 – Linear Algebra
▪ Matrix Multiplication, Eigenvalues, Eigenvectors

▪ Basic programming background in Python (an OK understanding of python

syntax is all that’s necessary – Geron’s textbook has excellent code examples)

▪ None of them are compulsory – easier to grasp the material if completed

 Course Project
▪ Groups of up to 3 members (3 is a hard maximum)
▪ Video Demo submission at the end of the semester, and in-person/online
presentation at the end of the semester
▪ 4-6 page Report due at semester end, IEEE format – must include link to Github
repo

▪ Potential Topics (few examples):

▪ Covid-19
▪ Computer Vision
▪ Natural Language Processing
▪ Reinforcement Learning
▪ Speech & Music Recognition
▪ Biomedical Imaging and Biosignals
 What is Machine Learning?
Tom Mitchell (1998): a computer program is
said to learn from experience E with respect
to some class of tasks T and performance
measure P, if its performance at tasks in T, as
measured by P, improves with experience E.

Example:
Task: Playing Checkers
Experience (data): games played by the
program (with itself)
Performance measure: winning rate Image from Tom Mitchell’s homepage
 Definition of Machine Learning
Arthur Samuel (1959): Machine Learning is the
field of study that gives the computer the ability
to learn without being explicitly programmed.

Photos from Wikipedia

 Traditional Programming

• Traditional Programming: writing a set of RULES to find

ANSWERS from DATA
 The ML Approach
Machine Learning: Use DATA and ANSWERS to learn the underlying set of RULES

Great for:

• Problems that require a lot of fine-

tuning or long list of rules
• Changing environments – ML
systems can ADAPT

• Getting insights from large amounts

of data

• Complex problems that yields no good

solution with traditional approach
 Deep Learning
▪ Subset of ML - loosely mimics
structure/function of human brain
▪ Unlike traditional ML, does not require
manual feature extraction
▪ Keeps getting better with more data
(typically)
▪ Computer Vision (CNN, GAN)
▪ Natural Language Processing (RNN,
LSTM)
▪ Automatic Speech Recognition (RNN)
 Summary – AI vs ML vs DL
▪ Subsets of each other
▪ 1950 – 1990: AI in the form of Expert systems (airplane
autopilot) and Games (checkers, chess)
▪ 1990- : Statistical Approaches with ML, busts AI winter
▪ 2010 - : Deep Learning revolutionizes CV, NLP among
other applications
▪ Narrow AI
▪ Systems can do a few defined things (such as playing
chess, or driving a car) as well, or better than humans
▪ Can’t do EVERYTHING a human being can do – yet
▪ AI is not “taking over the world” anytime soon
▪ Tell your uncles to relax and stop using Whatsapp
 What kind of ML system is it?
▪ Useful to classify ML systems based on the following criteria:
1. Does it require human supervision? 3. Does the system build a predictive model?
➢ Model-based Learning
➢ Supervised Learning
➢ Instance-based Learning
➢ Semisupervised Learning
➢ Unsupervised Learning
➢ Reinforcement Learning
• These are not exclusive – can be
combined
2. Can it learn incrementally on the fly?
➢ Online Learning
• e.g. Spam filter may learn on the fly
➢ Batch Learning with a deep neural network – online,
model-based, supervised learning
system
 Supervised Learning
▪ Training data fed to algorithm
includes the desired
answers/solutions (labels)
▪ Example algorithms:
▪ Linear Regression
▪ Logistic Regression
▪ SVM
▪ Decision Tree
▪ Neural Network
 Unsupervised Learning
▪ Training data is unlabeled
▪ System learns without direct human
supervision
▪ Widely used in:
▪ Clustering
▪ Anomaly detection
▪ Association mining
▪ Data preprocessing
▪ Example algorithms:
▪ K-means
▪ PCA
▪ SVD
▪ ICA
 Semisupervised Learning
▪ Partially labeled data
▪ Unsupervised learning used
to cluster similar data
together
▪ Human input taken to label
the clusters
▪ e.g. Google Photos will
cluster similar faces, and ask
the user if they are the
same person
 Reinforcement Learning
▪ The learning system (agent) can:
▪ Observe the environment
▪ Select and perform an action
▪ Get rewards/penalties as a result
▪ Learns what the best policy should be
▪ Policy defines what actions should be
chosen in a certain situation
▪ Very effective in controlled
environments (such as a game of chess)
▪ With the progress in deep learning,
increasingly used in more complex
tasks (such as driving the mars rover)
 Batch Learning vs Online Learning
▪ Batch Learning
▪ Not capable of learning after
deployment
▪ Must be retrained from scratch –
computationally expensive!
▪ Online Learning
▪ Can continue to learn after
deployment
▪ Can take advantage of parallel
computing – no down time
▪ Preferred choice in production
Example ML Task: Does money make people happy?

• Life Satisfaction data from OECD

• GDP per capita data from IMF

What relationship
can we infer between
life satisfaction and
GDP per capita from
the graph?
 Problems with Machine Learning
▪ 3 V’s of Big Data
▪ Volume, Variety, Velocity
▪ Problem #1: Training data!
▪ Insufficient quantity
▪ Nonrepresentative data
▪ Poor-quality data
▪ Problem #2: How “fit” is it?
▪ Overfitting data
▪ Underfitting data
▪ Problem #3: Which features should be used?
▪ Deep Learning automates feature selection
 Overfitting
▪ Most common problem in ML – do not overgeneralize!
▪ The polynomial model is better than the linear model on training
▪ How about testing?
 How to avoid overfitting
▪ Tip #1: REGULARIZATION – USE IT
▪ Constrain model to keep it simple – reduce risk of overfitting
▪ If you can stand on one leg, you’ll be able to stay balanced with two legs
▪ Hyperparameters – control level of regularization
▪ Tip #2: Get more training data, and reduce noise in it
 Model Evaluation
▪ How good is your model?
▪ Test it on new data – data not seen by the model ever before!
▪ Keep 80% for training, set 20% for testing
▪ NEVER go below 10% test data – better model is better than better
“accuracy”
▪ How to regularize?
▪ Keep a portion of training data held out for validation
▪ Alternatively, use cross-validation (many validation sets instead of one)
▪ Pick the hyperparameters that work best on validation for your model
on the test dataset
 Ratios
▪ A great model
▪ trained with 60% training data, 20% validation data, and 20% testing data
▪ An okay model
▪ trained with 70% training data, 15% validation data, and 15% testing data
▪ A barely-acceptable model
▪ trained with 80% training data, 10% validation data, and 10% testing data
▪ Models with worse ratios – hacks
▪ Unless there’s millions of instances in the dataset
▪ “No Free Lunch” theorem
▪ Only way to know for sure which model works best is to evaluate them
▪ Make reasonable assumptions about your data to select model