Module 1: Loss Functions & Optimization

Welcome to Module 1: Loss Functions & Optimization. In this module, we explore how an autonomous model learns by quantifying its own prediction errors into mathematical score surfaces and using calculus slopes to adjust its weights automatically.

Concepts in this Module

• Concept 01: Measuring Errors with Loss Functions (MSE & MAE)
  • The Everyday Problem: The robot shoots a game piece toward a target. How do we turn the difference between where it landed and where we aimed into an error score?
  • Code & Math: Mean Squared Error (MSE), Mean Absolute Error (MAE), and quadratic penalty curves.
• Concept 02: Cross-Entropy & Classification Loss
  • The Everyday Problem: The vision camera predicts probabilities for 3 game piece types. How do we heavily penalize a model that is confidently wrong?
  • Code & Math: Negative Log-Likelihood, Cross-Entropy Loss Loss = -ln(P_correct), and information surprise.
• Concept 03: Gradient Descent & Learning Rates
  • The Everyday Problem: How does an autonomous optimizer roll down the error bowl to find the best settings without overshooting?
  • Code & Math: Weight updates w_new = w - lr · ∇Loss, step sizes, learning rate tuning, and local minima.