Apache Spark Deep Learning Cookbook

There's more...

As seen in the preceding case, the error increased when the weight was increased but decreased when the weight was decreased. But this may not always be the case. The network uses the following graph to determine how to update weights and when to stop updating them:

• Let the weights be initialized to zero at the beginning of the first iteration. As the network updates its weights by increasing them from point A to B, the error rate begins to decrease.
• Once the weights reach point B, the error rate becomes minimal. The network constantly keeps track of the error rate.
• On further increasing the weights from point B towards point ...