Rerun the supervised learning neural network model given in Section 9.8.7 with a bottleneck containing just 2 dimensions several times. You should notice that the neural network often gets ‘stuck’;...

Rerun the supervised learning neural network model given in Section 9.8.7 with a bottleneck containing just 2 dimensions several times. You should notice that the neural network often gets ‘stuck’; sometimes it produces a model no better than random guessing (20% accuracy), other times it produces a model with 40% or 60% accuracy, and sometimes it produces a near perfect fit. What is causing this erratic behavior? Implement a fix and test that it produces a model that almost always produces a near-perfect fit of the response.