LeetCode 542 01 Matrix - Core Idea

You could run BFS from each $1$ cell to find the nearest $0$.

But if there are $n$ cells with $1$, you'd run BFS $n$ times. That's $O(n \times rows \times cols)$, which can time out.

Instead, reverse the problem. Push all $0$ cells into the queue at distance $0$. Then run BFS once. As you spread outward, you discover $1$ cells at distance $1$, then $1$ cells at distance $2$, and so on.

Why does this work? Because BFS explores in order of increasing distance. The first time you reach a $1$ cell from any $0$ cell, that's the shortest distance. You never need to revisit it.