Machine Learning Course Notes

Recognize the basic terminology and fundamental concepts of machine learning.
Understand the concepts of Supervised Learning models with a focus on recent advancements.
Relate the concepts of Neural Networks models of supervised Learning.
Discover Unsupervised learning paradigms of machine learning.
Understand the concepts of Reinforcement learning and Ensemble methods.

Explain the foundational concepts and prepare datasets for different Machine Learning models.
Identify and apply appropriate Supervised Learning models.
Design Neural Network models for given data.
Evaluate Machine Learning algorithms and perform Model Selection.
Devise unsupervised and Reinforcement learning models.

Predicting House Prices: Based on size, number of rooms, and location.
Stock Market Prices: Predicting current values based on historical data and market factors.
Email Filtering: Detecting spam vs. non-spam emails by analyzing words and sender details.
Image Recognition: Identifying faces and objects in images based on pixel data.
Text Summarization: Processing long documents to produce concise summaries.
Recommendation Systems: Suggesting movies/videos based on user history (e.g., YouTube, Netflix).
Chatbots: Designing chatbots for human interaction and query responses.
Autonomous Vehicles: Training self-driving cars for city navigation.
Medical Diagnosis: Assessing patient health status (sick vs. healthy).
Market Segmentation: Grouping markets by demographics and interests.
Game Playing: Developing AI to play games like chess.

Machine Learning Concepts: Covering supervised, unsupervised, semi-supervised learning, reinforcement learning, deep learning, and concept learning.
Feature Engineering: Methods for feature selection (Filter, Wrapper, Embedded), and feature normalization (min-max, z-score, etc.).
Dimensionality Reduction: Techniques include Principal Component Analysis (PCA) and Linear Discriminant Analysis (LDA).

Types of Learning: Focusing on regression and classification.
Regression Models: - Simple linear regression. - Multiple linear regression. - Cost Function: Evaluated using Gradient Descent. - Performance Metrics: Mean Absolute Error (MAE), Mean Squared Error (MSE), R-Squared, Adjusted R-Squared.
Classification Models: - Decision Trees (ID3, CART). - Naive Bayes. - K-Nearest Neighbors (KNN). - Logistic regression and Multinomial logistic regression. - Support Vector Machines (SVM).

Neural Network Representation: Overview of Artificial Neural Networks (ANN).
Activation Functions: Various types (e.g., Sigmoid, ReLU).
Learning Algorithms: Perceptron and Back Propagation.
Advanced Models: Introduction to Convolutional Neural Networks and Recurrent Neural Networks.
Classification Metrics: Usage of confusion matrix, precision, recall, accuracy, F-Score, ROC curves for performance evaluation.

Model Validation: Techniques include Cross Validation (Holdout, K-Fold, Stratified K-Fold, Leave-One-Out).
Model Optimization: - Bias-Variance Trade-off: Understanding overfitting vs. underfitting. - Regularization Techniques: Methods to mitigate overfitting.
Ensemble Methods: - Boosting. - Bagging. - Random Forest.

Clustering Techniques: K-means, K-Modes, K-Prototypes, Gaussian Mixture Models, Expectation-Maximization.
Reinforcement Learning: Strategies involving exploration vs. exploitation, non-associative learning principles, Markov decision processes, Q-learning.

Primary Text: "Machine Learning" by Tom M. Mitchell, -MGH
Further Reading: Kevin Murphy, "Machine Learning: A Probabilistic Perspective", MIT Press, 2012
Additional Text: R. S. Sutton and A. G. Barto, "Reinforcement Learning: An Introduction", MIT Press, 1998.
Reference Materials: - Trevor Hastie, Robert Tibshirani, Jerome Friedman, "The Elements of Statistical Learning", Springer 2009. - Christopher Bishop, "Pattern Recognition and Machine Learning", Springer, 2007. - Andrew Ng, "Machine Learning Yearning". - Jiawei Han and Micheline Kamber, "Data Mining - Concepts and Techniques", Morgan Kaufmann.

Definition: Machine Learning is the study of tasks where computers use data to make predictions and decisions without explicit programming.
Relationship: Distinction between Artificial Intelligence (AI) as the overarching domain, Machine Learning (ML) as a subset, and Deep Learning (DL) as a further specialized ML approach.

Data in ML: Utilizes both labeled and unlabeled data.
Types of Data: - Labeled Data: Data points accompanied by annotations or classifications. - Unlabeled Data: Raw data without associated tags.

Models: - Defined as sets of rules guiding predictions based on input data (e.g., a mathematical model of a phenomenon).
Algorithms: - Procedures for constructing models (e.g., Decision Trees).

Feature Selection: The process of identifying and selecting the most relevant features to improve model accuracy and efficiency. - Importance: Crucial in preventing overfitting and improving the model. - Methods: Include Filter, Wrapper, and Embedded methods.

Principal Component Analysis (PCA): Reduces dimensions by transforming to a new set of variables (principal components) that maintain the variance.
Linear Discriminant Analysis (LDA): Used for classification problems by projecting features into a lower-dimensional space, maximizing the separation between classes.

Focus on extracting patterns from data to make informed predictions and decisions in various applications.
Addressing the balance between model complexity and data representation through methods like feature selection, dimensionality reduction, and validation techniques.

Algorithms: Specific mathematical procedures for building models.
Evaluation Metrics: Used to assess the performance of algorithms and to guide adjustments in models for optimal performance.

Machine Learning is integral in automating decision-making processes in real-world applications, enhancing productivity, and uncovering insights from large sets of data. Continuous learning and adaptation are key components of successful machine learning applications.