Why Deep Learning Works Unreasonably Well [How Models Learn Part 3]

Transcript

In 1989, George Sabeno proved what's now known as the universal approximation theorem. If we take some complex function, for example, this really complicated border in the town of Barlay Hertok, these parts of the map are in Belgium and these parts are in the Netherlands. The universal approximation theorem guarantees that there exists a two-layer ... Read More