Fundamentals of ML
Fundamentals of Machine Learning
Author: Derbew Felasman (MSc), Debre Berhan University
Learning Objectives
Define Machine Learning (ML) and its significance.
Understand types of ML: Supervised, Unsupervised, Reinforcement Learning.
Recognize real-world applications of ML.
Explore the basic ML workflow.
Discuss key challenges and limitations of ML.
Understand relationships to other fields.
What is Machine Learning?
Definition: A field enabling computers to learn without explicit programming. (Arthur Samuel, 1959)
Key components:
Experience (E): Data the algorithm learns from.
Task (T): Problem the ML model is solved for.
Performance Measure (P): Metric for evaluating improvement.
ML vs. Traditional Programming
Traditional Programming
Characteristics:
Explicit instructions required.
Predictable output.
Rule-based decision making.
Well-defined problems.
Machine Learning
Characteristics:
Data-driven learning.
Probabilistic, uncertain predictions.
Pattern recognition in data.
Adaptability to new data.
Why Machine Learning Matters
Key Points:
Automates complex tasks and reduces manual work.
Uncovers hidden insights through data-driven decisions.
Drives innovation (e.g., medical diagnosis, self-driving cars).
Personalization of user experiences.
Economic impact through job creation and growth.
ML vs. Related Fields
Relations:
Artificial Intelligence (AI): Broader concept simulating human intelligence.
Data Science (DS): Encompasses broader processes including data collection and analysis.
Deep Learning (DL): Specialized ML using complex neural networks.
Types of ML Systems
Supervised Learning
Unsupervised Learning
Reinforcement Learning
Generative AI
Supervised Learning
Process:
Uses labeled input-output pairs to train algorithms.
Applications: Image recognition, spam filtering, weather forecasting.
Types of Algorithms:
Regression (predict continuous values) e.g., housing prices.
Classification (predict categorical outcomes) e.g., spam detection.
Unsupervised Learning
Definition:
Identifies patterns in data without labeled outcomes.
Techniques:
Clustering (e.g., K-Means).
Dimensionality reduction (e.g., PCA).
Reinforcement Learning
Definition:
Agents learn to make decisions through rewards and penalties.
Common Algorithms:
Q-Learning, Deep Q-Networks, Policy Gradient Methods.
The Machine Learning Workflow
Problem Definition
Data Collection
Data Preprocessing
Model Training
Evaluation
Deployment
Real-World Applications of ML
Sectors include:
Healthcare, Finance, Retail, Transportation.
Transforming industries—considered the new "electricity."
Challenges and Limitations of ML
Key Issues:
Data quality and bias.
Overfitting concerns.
Ethical issues (e.g., privacy, discrimination).
A Brief History of Machine Learning
1950s: Neural networks and pattern recognition inception.
1960s-1980s: Development of expert systems.
1990s: Rise of decision trees and support vector machines.
2000s-2010s: Advancements in deep learning.
Essential Math and Statistics for ML
Foundational Topics:
Linear Algebra, Calculus, Probability and Statistics.
Recap and Key Takeaways
Importance of ML, its types (Supervised, Unsupervised, Reinforcement).
Applications, workflow in ML, and ethical concerns.
Key Concepts: Data, Algorithms, and Models.