Add interactive lesson on overfitting vs underfitting

[LMK89]

Summary

Added a comprehensive educational lesson on overfitting and underfitting concepts with interactive visualizations and real polynomial fitting examples.

Key Changes

• New lesson file: 2026-08-22-overfitting-underfitting-c4cf9012.html

  • Complete standalone HTML lesson with embedded CSS and JavaScript
  • Vietnamese language content targeting beginners in machine learning

• Content sections:

  • Conceptual overview of bias-variance tradeoff with ASCII diagram
  • Four model states demonstrated with real least-squares polynomial fits (degree 1, 3, 7, and 7 with Ridge regularization)
  • Comparison table summarizing underfitting, good fit, overfitting, and regularized scenarios
  • Interactive demo allowing users to adjust regularization strength (λ) and observe MSE changes
  • Real-world examples of good vs bad datasets
  • Practice exercise with answer reveal
  • 4-question quiz with immediate feedback

• Interactive features:

  • Canvas-based visualization of fitted curves with progressive animation
  • Real computed coefficients from polynomial fitting (not fabricated data)
  • Live regularization strength slider with precomputed Ridge regression results
  • Theme toggle (light/dark mode) with system preference detection
  • Intersection Observer for scroll-triggered animations
  • Responsive design with mobile support

• Updated README: Added entry for the new lesson in the course index table

Implementation Details

• Uses real polynomial fitting data (8 training points, 4 test points) with actual least-squares coefficients
• Canvas rendering with device pixel ratio support for crisp graphics
• Precomputed Ridge regression results at λ = 0, 0.001, 0.1, 1, 5 for the demo
• CSS custom properties for theming with automatic dark mode detection
• Vanilla JavaScript (no external dependencies) for interactivity
• Accessible form controls with proper ARIA labels

https://claude.ai/code/session_01S64DiCZUs6hmhhxSBjVfAH