Interactive Loss Calculator
Choose what the model predicts:
mat
chair
table
floor
elephant
Cross-Entropy Loss:
L = -log(p) = -log(0.8) = 0.223
L = -log(p) = -log(0.8) = 0.223
Current Prediction Scenario
Prediction: "mat"
Actual correct word: "mat"
✅ Correct prediction!
80%
Loss: 0.223
Try Different Scenarios
Click different words above and adjust confidence to see how loss changes:
- High confidence + correct = Low loss (good!)
- High confidence + wrong = High loss (bad!)
- Low confidence + correct = Medium loss
- Low confidence + wrong = Medium loss
Key Insights
- Lower loss means better predictions
- Cross-entropy heavily penalizes confident wrong predictions
- The model learns by trying to minimize this loss
Loss Behavior Visualization
Example Scenarios
Why Cross-Entropy?
- Smooth gradient: Provides clear direction for improvement
- Penalizes confidence: Wrong predictions with high confidence get heavily penalized
- Probabilistic: Works naturally with softmax probability outputs
- Information theoretic: Measures how "surprised" the model is by the correct answer