Probabilistic and Stochastic Neural Networks

As discussed earlier, coming up with a parameterized model is often the hard part in machine learning. Up until this point, we have tried to specify causal models with an explicit analytic expression for its components to specify a multinomial density function. In Chapter 5, we saw how we could use neural networks as a tool to specify complex functional models. The question now is how we can reconcile neural networks with the probabilistic view of modeling. Here, we’ll touch on three aspects of this discussion:

• Generalistic versus specific models
• Representing probability functions
• Stochastic models