trendingAI & ML — LAB
Gradient Descent Lab
Set a starting point, a target, and a learning rate, then step through gradient descent and watch the trail of steps on the loss curve. This is the same mechanism covered in Gradient Descent & Loss Functions, built as a free-form workspace instead of fixed presets.
How to Use This Lab
- Drag the starting w, target, and learning rate sliders to whatever values you want.
- Click “Step” repeatedly to watch w move toward the target — each step is added to the trail on the curve, so you can see the whole path, not just where it ends up.
- Try a very high learning rate and step a few times — watch the trail overshoot the target and grow farther away instead of closer, which is divergence.
- Reset returns to your chosen starting point without changing target or learning rate.
The Lab
Want the full explanation — batch vs. stochastic vs. mini-batch descent, momentum, and the Adam optimizer? Read Gradient Descent & Loss Functions →