The other path returns

Learning from examples was not invented after expert systems faded. Researchers had explored small networks of connected units since the 1940s.

Interest rose, fell, and returned. Researchers had explored ways to send error corrections backwards through a network before the 1980s. In 1986, Rumelhart, Hinton, and Williams published a clear demonstration of the method on networks with hidden layers. The paper helped make more ambitious learning experiments practical and visible.

The names and mechanics can wait. The larger change was a move away from writing every decision as a rule. Researchers were finding ways for useful patterns to emerge from examples.

The next module returns to the beginning of that path and builds it one small piece at a time.

The other path returns

Learning from examples was not invented after expert systems faded. Researchers had explored small networks of connected units since the 1940s.

The names and mechanics can wait. The larger change was a move away from writing every decision as a rule. Researchers were finding ways for useful patterns to emerge from examples.

The next module returns to the beginning of that path and builds it one small piece at a time.