Gradient Optimization
Like a blindfolded climber finding the lowest valley. The gradient acts as a mathematical compass, guiding parameters to minimize error.
W
new
= W
old
- α ∇L