11.3b Training and backprop labs

Source: Lab ideas from Computational Physics (Landau, Páez, Bordeianu), §11.2.3–11.2.5. Original drills. Continues 11.3.

Explorer

Pure Python. No NumPy.

Formula refresh

Warm up: pick the matching equation among lookalikes (same walk as the theory lesson).

Labs

Activity: $\mathcal{L}=\mathrm{mean}\,(y^c-y^p)^2$

With $y^c=[1,0,0,1]$ and $y^p=[0,0,0,0]$, print $\mathcal{L}$ with 10 decimals.

No input.

Write-ups

0/3 solved

11.3b.W1

Using “backprop”, “chain”, and “gradient”, explain:

Why does backpropagation multiply several partial derivatives when updating an early weight like w₁?

Minimum 3 words. Your answer must include: “backprop”, “chain”, “gradient”.

0 words/ 3 min

Meet the writing requirements, then you can open the model answer.

Next: 11.4 A graphical deep net.