Training

• This was really popular in the 80s and 90s, and is still used today.
• You need to train the system with data where you know the answers.
• This is perfect for a two fold test.
• One question is how long you train for.
• Typically this is done until there is only a certain amount of error.
• The real question is not can it find a weight setting that matches the training data, but can it find a weight setting that matches the test data.