Lecture Notes: CS 221 - Artificial Intelligence

Course Introduction

• Instructor: Percy Liang
• Co-Instructor: Dorsa
• Course focuses on AI, taught through theory and practical applications.

Teaching Team

• Diverse interests among teaching assistants (TAs):
  • Natural Language Processing
  • Machine Learning
  • Computer Vision
  • Reinforcement Learning
  • AI Education
• Students encouraged to utilize TA expertise for projects.

Announcements

• Weekly sections
  • Cover review and advanced topics.
  • Python and probability overview this Thursday.
• Homework 1 is posted and due next Tuesday by 11 PM.
• Submissions via Gradescope, code posted on Piazza.

AI's Importance and History

• AI is now widely recognized as crucial.
• Success stories include AI in games, face recognition, and medical imaging.

Early AI Developments

• Dartmouth Conference (1956): Sparked AI development.
• Programs developed for games and theorem proving.
• First AI Winter: Due to over-promising and lack of computing power.

AI in the 70s and 80s

• Focus shifted to expert systems
• Real industrial impact but led to second AI winter due to limitations and over-promising.

Neural Networks

• 1943: McCullough and Pitts developed theory of artificial neural networks.
• 1980s: Backpropagation rediscovered, leading to interest in multi-layer networks.
• 2012: Deep learning took off with AlexNet.

AI's Dual Perspectives

• AI as Agents: Recreating human-like intelligence.
• AI as Tools: Using AI to enhance human capabilities.
• Challenges include biases in data and societal implications of AI applications.

Course Content Overview

• Modeling, Inference, Learning Paradigm: Structured approach to solving AI problems.
• Topics to be covered:
  • Machine Learning
  • Reflex Models
  • State-Based Models
  • Variable-Based Models
  • Logic-Based Models

Course Logistics

• Prerequisites: Programming, discrete math, probability.
• Goals: Provide AI tools and improve proficiency in math and programming.
• Coursework: 8 homeworks, exams, a project.

Project Details

• Group work (up to 3 people).
• Involves brainstorming, progress reports, and a poster session.
• Mentorship from CAs available.

Policies

• Submissions via Gradescope.
• 7 total late days allowed for assignments.
• Communication through Piazza.
• Honor code emphasized: No copying or sharing of code and solutions.

Technical Details: Optimization

• Discrete Optimization: Finding best discrete object (dynamic programming).
• Continuous Optimization: Finding best vector of real numbers (gradient descent).

Dynamic Programming Example

• Problem: Calculate edit distance between two strings.
• Solution: Use recurrence relations and memoization to optimize.

Gradient Descent Example

• Problem: Perform linear regression to fit a line.
• Solution: Use gradient descent to minimize error function.

Conclusion

• Next class will start on machine learning.
• Focus on moving complexity from code to data through machine learning paradigms.