Result showing that sufficiently large neural networks with non‑linear activations can approximate any continuous function on compact domains to arbitrary accuracy.