🧠 Loss Functions — ML Simulator
Loss Function

Compare how MSE, MAE & Huber penalise errors.

Outlier Error3.0
Huber δ1.0
At Error = 3.0
MSE—
MAE—
Huber—
Log-Cosh—
Key Insight
MSE squares errors — outliers get heavily penalised. MAE is robust but not smooth at 0.
Regression Setup

Drag outlier slider. See how each loss responds.

Outlier Magnitude2.0
Noise σ0.3
Loss Values
MSE—
MAE—
Huber—
RMSE—
Legend
● Data● Outlier
— MSE fit— MAE fit
Loss Type
Predicted Prob p0.7
True Label1
Loss at p=0.70, y=1
Cross-Entropy—
Hinge—
Focal (γ=2)—
Binary Acc—
Use Case
Cross-Entropy: ideal for probabilities — penalises confident wrong predictions heavily.
Optimizer Setup
Learning Rate0.10
Momentum β0.00
Start θ-3.0
Training Log
Epoch0
θ (weight)—
Loss—
Gradient—
Adjust sliders and press Step or Run...