The atomic operation of neural network training. Update parameters by moving in the **negative gradient direction**.
θ_new[i] = θ_old[i] - lr × gradient[i]gradient_step([1.0, 2.0], [0.1, 0.2], lr=0.1) → [0.99, 1.98]
- SGD: gradient from one sample
- Mini-batch: gradient averaged over a batch
- Adam: adaptive learning rate per parameter
Round to **5 decimal places**.
Similar Problems
Test Cases (2 visible · 1 hidden)
Case 1: Basic step
Input: gradient_step([1.0,2.0],[0.1,0.2],0.1)
Expected: [0.99, 1.98]
Case 2: Single param
Input: gradient_step([0.0],[1.0],0.01)
Expected: [-0.01]
⌘↵ Run · ⌘⇧↵ Submit