Machine Learning from Scratch & Its Privacy Implications

Begin Date: 8th July - Ongoing

This course aims to teach the fundamentals of machine learning from scratch while also addressing the privacy implications at each step of the process. The curriculum is designed to provide a comprehensive understanding of machine learning techniques and their privacy considerations.

Table of Contents

• Trainers
• Moderators
• Prerequisites
• Course Outline

Trainers

Ahmad Jajja

Asjad Ali

Zartashia Afzal

Moderators

Mahnoor Malik

Muhammad Arham

Sheraz Anwar

Sikander Nawaz

Prerequisites

• There are no prerequisites to join this course. You'll learn from zero to advanced level.

Course Outline

Module 1: Introduction to Machine Learning

• What is Machine Learning?
• Applications of Machine Learning
• Machine Learning Development Life Cycle (MLDLC)
• Importance of Machine Learning in the Generative AI Era (Optional)
• Introduction to Differential Privacy (DP)
  • Definition and Importance
• Class 1 Video Link
• Class 2 Video Link
• Class 3 Video Link
• Class 4 Video Link
• Class 5 Video Link

Module 2: Python for Machine Learning (Optional)

• Introduction to Python Programming (if needed)
• Libraries for Data Analysis: Pandas, NumPy
• Introduction to Privacy Libraries in Python
  • Libraries for Implementing Differential Privacy: PySyft, PyTorch Opacus
• Class 6 Video Link
• Class 7 Video Link
• Class 8 Video Link

Module 3: Data Preprocessing and Feature Engineering

• Data Analysis and Preprocessing Techniques
• Data Cleaning: Handling Missing Data, Categorical Features, Outliers
• Data Visualization with Seaborn and Matplotlib
• Feature Engineering: Feature Transformation, Selection, Construction, and Extraction
• Dimensionality Reduction with PCA (Principal Component Analysis)
• Privacy-Preserving Data Preprocessing
  • Anonymization Techniques
  • Privacy Risks in Data Preprocessing
• Class 9 Video Link
• Class 10 Video Link
• Class 11 Video Link
• Class 12 Video Link
• Class 13 Video Link
• Class 14 Video Link
• Class 15 Video Link
• Class 16 Video Link
• Class 17 Video Link
• Class 18 Video Link
• Class 19 Video Link
• Class 20 Video Link
• Class 21 Video Link

Module 4: Machine Learning Fundamentals

• Learning Approaches: Batch vs Online, Model-based vs Instance-based
• Types of Machine Learning: Supervised, Unsupervised, Semi-Supervised, Reinforcement Learning
• Privacy Risks in Different Learning Approaches
  • Supervised Learning: Risks of Label Leakage
  • Unsupervised Learning: Risks in Clustering and Association

Module 5: Supervised Learning Algorithms

• Introduction to Supervised Learning
• Regression vs. Classification
• Regression Algorithms: Simple Linear Regression, Multilinear Regression, Polynomial Regression (with applications like house price prediction)
• Classification Algorithms: Decision Trees (Decision Tree Classifier, Random Forest), K-Nearest Neighbors (KNN), Naive Bayes, Support Vector Machines (SVM)
• Differential Privacy in Supervised Learning
  • Noise Addition in Regression Models
  • Privacy-Preserving Decision Trees

Module 6: Model Evaluation and Optimization

• Regression and Classification Metrics
• Imbalanced Data in Machine Learning
• Underfitting vs Overfitting
• Ensemble Methods: Bagging, Boosting
• Hyperparameter Tuning
• Privacy-Preserving Model Evaluation
  • Metrics for Assessing Privacy Risks
  • Differential Privacy in Model Optimization

Module 7: Model Interpretation and Deployment

• Model Interpretability and Explainable AI (XAI)
• Model Deployment with Flask (or similar framework)
• Privacy Concerns in Model Interpretation
  • Risks of Exposing Sensitive Information through Interpretability
• Privacy-Preserving Model Deployment
  • Secure Multi-Party Computation for Model Serving

Module 8: Unsupervised Machine Learning

• Unsupervised Learning goals (e.g., clustering, dimensionality reduction)
• Common applications (e.g., customer segmentation, anomaly detection)
• Specific algorithms:
  • K-Means Clustering
  • Hierarchical Clustering
  • Principal Component Analysis (PCA) for dimensionality reduction
  • Association Rule Learning (e.g., Apriori algorithm)
• Privacy Risks in Unsupervised Learning
  • Privacy-Preserving Clustering Techniques
  • Anonymization in Dimensionality Reduction

Additional Topics (Optional)

• Unit Testing, Feature Store, Model Registries
• Containerization with Docker
• Introduction to Time Series Analysis
• Machine Learning Competitions