$$ w_{new} = w - \alpha \frac{\partial Loss}{\partial w} $$
Values
Learning Rate
Gradient Descent Explorer
Iteration:
0
Parameter w:
0.00
Loss L(w):
0.00
Gradient:
0.00
How it works
- Curve: The "Loss Landscape". We want to find the bottom (minimum loss).
- Gradient: The slope. Tells us which way is "down".
- Alpha (α): The step size.
- Small: Safe but slow.
- Good: Fast convergence.
- Large: Overshoots, unstable.