Course Goal: To equip students with machine learning skills to build meaningful applications and possibly transform industries through ML tools and techniques.
Class Logistics
Enrollment: Over 800 students enrolled; sessions recorded and available online via SCPD.
Prerequisites:
Basic computer science concepts: Big O notation, queues, stacks, binary trees.
Probability: Random variables, expected value, variance.
Linear algebra: Matrices, vectors, eigenvectors.
Tools: Transition from MATLAB/Octave to Python/NumPy for assignments.
Honor Code: Do homework individually after possible discussions with friends. Written solutions must reflect individual understanding.
Course Structure
Lectures: Mondays and Wednesdays
Discussion Sections: Fridays (optional but beneficial).
Course Communication: Use Piazza for most queries and student interaction.
Assignments: Four homeworks, a project proposal, and a final project. Key dates on the course website.
Midterm: Take-home format instead of timed.
Machine Learning Goals
Supervised Learning: Focus on learning functions mapping from input (X) to output (Y).
Types: Regression (continuous Y) and Classification (discrete Y).
Examples: Predicting house prices, diagnosing cancer from tumor sizes.
Methods: Linear regression, logistic regression, support vector machines (SVM).
Application Example
Autonomous Driving: An example using supervised learning where a neural network learns to drive by imitating human steering based on camera images.
Strategic Application of ML
Learning Theory: Systematic approaches to effectively debug and improve ML algorithms using concepts from software engineering.
Machine Learning Strategy: Focusing on the most promising techniques to quickly identify which approaches will work, inspired by principles from optimizing code and software engineering.
Deep Learning
Focus: Neural networks and deep learning concepts, progressing in both theory and application, specialty explored deeply in CS230.
Unsupervised Learning
No Labels: Finding interesting patterns in data without predefined labels.
Examples: K-means clustering, market segmentation, social network analysis, genetic data analysis.
Applications: Clustering data, understanding large datasets, cocktail party problem (separating overlapping sounds).
Reinforcement Learning
Behavioral Training: Learning based on feedback from actions to optimize future behaviors.
Examples: Training the Stanford autonomous helicopter, optimizing robot movements.
Modern Application: Google’s AlphaGo, robotics, logistics.
Class Projects
Group Work: Form study groups and project teams (up to three members, exceptions for larger groups).
Finding Projects: Brainstorming to apply ML to interesting problems; past projects available on the course website for inspiration.
Additional Information
Emails and Office Hours: Use Piazza for general queries; course staff available via email for private concerns.
Course Adjustments: Adapting syllabus and tools to latest developments in ML.
Encouragement
Stanford Community: First class for many students; opportunity to network and collaborate right from the start.
Future Prospects: Significant demand for ML skills across industries; course aims to prepare students to lead and innovate in various sectors.