Add slide explanining our vector + vector and matrix + vector conventions

slides/01/01.md

@@ -265,6 +265,23 @@ class: section
 - $∇_{→x} f(→x)$: gradient of $f$ with respect to $→x$, i.e.,
   $\left(\frac{∂f(→x)}{∂x_1}, \frac{∂f(→x)}{∂x_2}, \ldots, \frac{∂f(→x)}{∂x_n}\right)$
 
+---
+# Linear Algebra Conventions & Broadcasting
+
+## Vector Addition
+
+Linear algebra distinguishes between column vectors ($N\times 1$ matrices) and row vectors ($1\times N$ matrices), often treating the addition of a row and a column as undefined.
+
+~~~
+However, in deep learning, vectors that are outside of multiplication are treated simply as 1D arrays. Therefore, vector + vector is valid if their lengths match.
+~~~
+
+## Matrix-Vector Broadcasting
+
+When adding a matrix and a vector, we adopt broadcasting conventions common in NumPy and PyTorch. The vector is implicitly copied to match the matrix shape.
+
+![w=40%,h=center](broadcasting.png)
+
 ---
 section: Random Variables
 class: section

slides/01/broadcasting.png.ref

@@ -0,0 +1 @@
+Figure 2 of NumPy Broadcasting, https://numpy.org/doc/stable/user/basics.broadcasting.html