1 of 59

Introduction to Machine Learning

Based on:

Hands-On Machine Learning with Scikit-Learn and PyTorch (Aurélien Géron)

Ibrahim Said Ahmad, PhD

Week 1: Introduction to Machine Learning

2 of 59

Meet Your Instructor: Ibrahim Said Ahmad, PhD

Bridging AI research with practical education to empower the next generation of computing professionals

3 of 59

Academic Leadership and Global Experience

Assistant Professor

UW–Stevens Point

2025 - Present

Leading curriculum development and student mentorship in the Department of Computing and New Media Technologies.

Postdoctoral Research Fellow

Northeastern University

2023 - 2025

Conducting advanced research at the Institute for Experiential AI on NLP and computational social science.

Lecturer (~Assistant Professor)

Bayero University Kano

2014 - 2023

Teaching AI, Machine Learning, and Database Systems while mentoring graduate research projects.

4 of 59

Research at the Frontier of AI and Society

Natural Language Processing

Specializing in NLP techniques that address real-world challenges in understanding and processing human language.

Computational Social Science

Applying computational methods to understand social phenomena through sentiment analysis, hate speech detection, and emotion analysis.

Low-Resource Languages

Dedicated to addressing linguistic and cultural disparities in AI systems through dataset creation and cultural adaptation.

5 of 59

Award-Winning Research and Global Initiatives

Competitive Grants

    • Over $400,000 from Google Research, Lacuna Fund, and Oracle

Research Awards

    • Best Paper, ACII 2025
    • Best Resource Paper Award, ACL 2025 (BRIGHTER)
    • SemEval Best Task Award, ACL 2025
    • Honorable Mention, SemEval 2024

Global Initiatives

    • Co-founder, HausaNLP
    • Co-founder, ArewaDataScience

6 of 59

Hands-On Education

Expert instruction in Artificial Intelligence, Machine Learning, and Programming with emphasis on practical application.

Mentorship Through Research

Guiding students through research projects and real-world problem solving to develop critical thinking skills.

Student Success

Committed to empowering students with technical skills and knowledge needed for the modern AI landscape.

Teaching Philosophy: Bridging Theory and Practice

7 of 59

About the Course

8 of 59

What You'll Learn

Machine learning fundamentals

Core concepts, terminology, and ML categories

Python ML workflows

Data preprocessing, feature engineering, and complete pipelines

Classical ML algorithms

Classification, regression, clustering, and ensemble methods

Model evaluation

Interpret outputs, use metrics, and design ethical AI systems

9 of 59

Course Structure

20 points

Assignments

5 assignments on AI concepts

40 points

Course Project

Build an AI application

30 points

Exams & Quizzes

4 quizzes + midterm + final

10 points

Participation

Active engagement in class

Total: 100 points

10 of 59

Academic Integrity

All work must be original and individual

Unless collaboration is explicitly allowed in assignment instructions

AI Use Policy

  • Unauthorized use of generative AI is not permitted
  • Microsoft Copilot may be authorized for specific assignments
  • Always document AI use when permitted

Using third-party resources

Properly cite any code or materials you incorporate. Failure to cite is academic misconduct.

11 of 59

Late Submission Policy

25%

Late Penalty

25% of total points deduction per day late submission penalty

0

Points Awarded

For submissions more than 4 days late

Request extensions BEFORE the due date for legitimate reasons

12 of 59

Course Topics

Weeks 1-4

ML landscape & end-to-end projects

Weeks 5-6

Classification & model evaluation

Weeks 7-8

Linear models & regularization

Week 9

Support Vector Machines

Weeks 10

Decision Trees

Weeks 11

Ensemble methods (Random Forest, Boosting)

Weeks 12-13

Dimensionality reduction & clustering

Weeks 14-16

Integration, final project & presentations

13 of 59

Tools & Keys to Success

Tools We'll Use

Keys to Success

Attend class regularly

Start assignments early

Ask questions when stuck

Practice coding between classes

Python

Programming language

Google Collab

Free development environment

ML Libraries

NumPy, Pandas, Scikit-Learn, Matplotlib

14 of 59

  • Machine learning powers:
    • Search engines
    • Recommendation systems
    • Voice assistants
    • Chatbots
  • ML is no longer “behind the scenes”

Why This Course?

Goal: learn how ML systems are designed, trained, and evaluated

15 of 59

Learning Objectives

By the end of this week, you should be able to:

Define machine learning formally

Explain when ML is preferable to traditional programming

Classify ML systems

Describe a typical ML workflow

Identify common ML pitfalls

Understand model evaluation basics

16 of 59

  • Machine learning is the science (and art) of programming computers so they can learn from data.
  • Here is a slightly more general definition:

[Machine learning is the] field of study that gives computers the ability to learn

without being explicitly programmed.

—Arthur Samuel, 1959

  • And a more engineering-oriented one:

A computer program is said to learn from experience E with respect to some task

T and some performance measure P, if its performance on T, as measured by P,

improves with experience E.

—Tom Mitchell, 1997

What Is Machine Learning?

17 of 59

A program learns from:

    • Experience (E)
    • With respect to Task (T)
    • As measured by Performance (P)

Formal Definition (Tom Mitchell)

Example:

    • Task: spam detection
    • Experience: labeled emails
    • Performance: accuracy

18 of 59

Key ML Terminology

Training Set

Training instance (sample)

Feature

Model

dataset

Accuracy (example metric)

Evaluation

19 of 59

  • Downloading data ≠ learning
  • Storing information ≠ intelligence

Learning requires:

  • Task
  • Feedback
  • Performance improvement

What ML Is Not

20 of 59

Traditional programming struggles when:

    • Rules are complex
    • Environments change
    • No known algorithm exists
    • Data is large and noisy

Why Use Machine Learning?

21 of 59

  • Hard-coded rules
  • Manually analyze patterns
  • Write detection rules
  • Maintain forever

Figure 1-1. The traditional approach

Traditional Programming Approach

22 of 59

  • Learn patterns from data
  • Shorter code
  • Easier maintenance
  • Better adaptability

Figure 1-2. ”The machine learning approach

Machine Learning Approach

23 of 59

  • ML systems retrain automatically
  • Detect new patterns without manual updates

Figure 1-3. Automatically adapting to change

Adapting to Change

24 of 59

Examples:

  • Speech recognition
  • Image recognition
  • Language translation
  • Why ML?
    • No scalable rule-based solution exists

ML for Hard Problems

25 of 59

  • Discover hidden patterns
  • Reveal correlations
  • Enable data mining

Figure 1-4. Machine learning can help humans learn

ML Can Help Humans Learn

26 of 59

ML is useful for:

  • Complex rule systems
  • Changing environments
  • Unknown algorithms
  • Insight discovery

Summary: Why ML?

27 of 59

  • Image classification
  • Tumor detection
  • Text classification
  • Fraud detection
  • Recommendation systems
  • Games and robotics

Examples of ML Applications

28 of 59

Three major axes:

    • Training supervision
    • Batch vs online learning
    • Instance-based vs model-based learning

How Can We Classify ML Systems?

29 of 59

  • Labeled data
    • Classification
    • Regression

Examples:

  • Spam detection
  • House price prediction

Figure 1-5. A labeled training set for spam classification

Supervised Learning

30 of 59

  • Predict numeric values

  • Output is continuous

Figure 1-6. A regression problem: predict a value, given an input feature

Regression Example

31 of 59

  • No labels
  • Discover structure
  • Patterns emerge automatically

Examples:

  • Clustering
  • Visualization
  • Anomaly detection

Unsupervised Learning

32 of 59

  • Group similar instances
  • No predefined categories

Figure 1-7. Clustering

Clustering

33 of 59

  • Reduce complexity
  • Preserve structure

Figure 1-8. Example of a t-SNE visualization highlighting semantic clusters

Visualization & Dimensionality Reduction

34 of 59

  • Identify unusual behavior
  • Fraud detection
  • Manufacturing defects

Figure 1-9. Anomaly detection

Anomaly Detection

35 of 59

  • Few labels
  • Many unlabeled examples
  • Combines supervised + unsupervised learning

Figure 1-10. Semi-supervised learning with two classes

Semi-Supervised Learning

36 of 59

  • Labels generated automatically
  • Foundation of modern deep learning
  • Pretraining → fine-tuning

Figure 1-11. Self-supervised learning example: input (left) and target (right)

Self-Supervised Learning

37 of 59

  • Agent
  • Environment
  • Actions
  • Rewards
  • Policy

Figure 1-12. Reinforcement learning

Reinforcement Learning

38 of 59

  • Train once on full dataset
  • Offline learning
  • Requires retraining to adapt

Batch Learning

39 of 59

  • Incremental updates
  • Mini-batches
  • Continuous learning

Figure 1-13. In online learning, a model is trained and launched into production, and then it keeps learning as new data comes in

Online Learning

40 of 59

  • Datasets too large for memory
  • Learn chunk by chunk

Figure 1-14. Using online learning to handle huge datasets

Out-of-Core Learning

41 of 59

Instance-Based Learning

  • Memorize examples
  • Similarity-based predictions

Figure 1-15. Instance-based learning: in this example we consider the class of the three nearest neighbors in the training set

Instance-Based Learning

42 of 59

  • Learn parameters
  • Build predictive model

Figure 1-16. Model-based learning

Model-Based Learning

43 of 59

  • Study the data
  • Select a model
  • Train the model
  • Make predictions

A Typical ML Workflow

44 of 59

  • Input: GDP per capita
  • Output: Life satisfaction
  • Model: linear regression

Figure 1-17. Do you see a trend here?

Example: GDP vs Life Satisfaction

45 of 59

  • Choose model type
  • Define parameters
  • Choose loss function

Figure 1-18. A few possible linear models

Model Selection

46 of 59

Training

  • Minimize loss
  • Fit training data

Figure 1-19. The linear model that fits the training data best

Training & Inference

Inference

  • Apply trained model
  • Predict on unseen data

47 of 59

Two big categories:

  • Bad data
  • Bad models

Main ML Challenges

48 of 59

  • ML needs lots of examples
  • Especially for complex tasks

Figure 1-21. The importance of data versus algorithms

Insufficient Training Data

49 of 59

  • Sampling bias
  • Nonresponse bias
  • Leads to poor generalization

Figure 1-22. A more representative training sample

Non-Representative Data

50 of 59

  • Noise
  • Missing values
  • Outliers
  • Irrelevant features

Poor-Quality Data

Image source: https://markkolier.com/wp-content/uploads/2016/11/bad-data1.png

51 of 59

Overfitting

  • Model too complex
  • Learns noise
  • Poor generalization

Overfitting and Underfitting

Underfitting

  • Model too simple
  • Misses patterns
  • Poor training performance

Balance

52 of 59

  • Constrain model complexity
  • Reduce overfitting

Figure 1-24. Regularization reduces the risk of overfitting

Regularization

53 of 59

  • Training set: learn
  • Test set: evaluate
  • Measure generalization error

Training vs Test Set

54 of 59

  • Tune hyperparameters
  • Avoid test set leakage

Figure 1-25. Model selection using holdout validation

Validation Set

55 of 59

  • Training data ≠ production data
  • Use train-dev set

Figure 1-26. Train-Dev Strategy

Data Mismatch

56 of 59

  • No universally best model
  • Assumptions matter
  • Evaluation is essential

No Free Lunch Theorem

57 of 59

  • ML learns from data
  • Many system types exist
  • Data quality matters
  • Evaluation is critical

Big Picture Summary

58 of 59

Next week:

  • End-to-end ML project
  • First hands-on coding
  • Scikit-Learn workflow

What’s Next?

59 of 59

  • Read Chapter 1
  • Review terminology
  • Install Python / Jupyter / Colab

Reading & Homework