Home▸AI & Machine Learning▸Loss Function Hyperparameters Guide

📉 Loss Function Hyperparameters Guide

Watch a marble descend a live 3D loss-landscape while you switch between MSE, MAE and Huber loss, add an outlier, and tune the learning rate, Huber delta and L2 regularization.

AI & Machine Learning3DAdvanced60 FPS

A live 3D loss landscape for a tiny linear model, where a marble performs gradient descent while you switch between MSE, MAE and Huber loss and tune the hyperparameters that shape each one.

🔬 What It Demonstrates

The surface height at each point is the loss for those weights. Switching loss function reshapes the bowl live; adding an outlier shows why MSE distorts badly while MAE and Huber stay comparatively stable.

🎮 How to Use

Pick a loss function, tune the learning rate, Huber delta and L2 regularization, then toggle the outlier point and watch the marble's descent path change on the reshaped surface.

💡 Did You Know?

Huber loss's δ hyperparameter directly controls the trade-off between MSE's smooth convergence and MAE's outlier robustness — a single knob blending two loss families.

⚙ Under the hood

Watch a marble descend a live 3D loss-landscape while you switch between MSE, MAE and Huber loss, add an outlier, and tune the learning rate, Huber delta and L2 regularization.

machine learningloss functionshyperparametersoptimizationneural networksgradient descentThree.js

3D · Three.js / WebGL renderer · 60 FPS target · runs fully client-side, no install