CS5720 - Week 2
Slide 22 of 40

Loss Functions: Measuring Errors

What is a Loss Function?

A loss function (also called cost function) quantifies how wrong our neural network's predictions are compared to the actual correct answers.
Why do we need it?

• To know how wrong we are
• To have a single number to minimize
• To guide the learning process
• To compare different models
🎯 Think of it like...
A loss function is like a GPS telling you how far you are from your destination. The bigger the loss, the further you are from the correct answer!

Common Loss Functions

  • 📊
    Regression Loss
    For predicting continuous values (prices, temperatures, scores)
  • 🏷️
    Classification Loss
    For predicting categories (cat/dog, spam/not spam)
  • 🔧
    Specialized Loss
    For specific tasks (object detection, segmentation)
Remember:
Different problems need different loss functions. Choosing the right one is crucial for successful training!

Loss Function in Action

Excellent Prediction
0.01
Predicted: 4.99
Actual: 5.00
Almost perfect!
Okay Prediction
0.25
Predicted: 4.50
Actual: 5.00
Getting there...
Poor Prediction
4.00
Predicted: 3.00
Actual: 5.00
Needs improvement!
Goal: Minimize the loss function to make better predictions!
← Previous Next →
Prepared by Dr. Gorkem Kar