Machine Learning and the Bigdata Paradigm
157
Fig. 4 Transfer function, also called the activation function, can be any function that is differentiable
and has a finite output range [12]. The table shows a set of possible activation functions and their
plots
1.2 The Difference Boosting Neural Network
While ANN is a great idea and a breakthrough technology, it is not guaranteed to give
the best results. This is mostly because there could be multiple minima in the error
surface and instead of reaching the global minima, the weights might get trapped
elsewhere that may give sub-optimal performance. Moreover, the iterative training
process may take a long time to converge, thus adding to the computational overhead.
A more profound model is the Bayesian formalism that relies on likelihood and
circumstantial knowledge for making efficient decisions. The likelihood is the experiential knowledge about the fraction of times a feature observation has resulted in
a particular event. For example, clouds may cause it to rain and the likelihood is the
fraction of times observation of clouds resulted in rain. Circumstantial knowledge is
also a consideration in Bayesian formalism. It is called the prior. It is the background
information about the probability of the event to happen in a given situation. For
157
Fig. 4 Transfer function, also called the activation function, can be any function that is differentiable
and has a finite output range [12]. The table shows a set of possible activation functions and their
plots
1.2 The Difference Boosting Neural Network
While ANN is a great idea and a breakthrough technology, it is not guaranteed to give
the best results. This is mostly because there could be multiple minima in the error
surface and instead of reaching the global minima, the weights might get trapped
elsewhere that may give sub-optimal performance. Moreover, the iterative training
process may take a long time to converge, thus adding to the computational overhead.
A more profound model is the Bayesian formalism that relies on likelihood and
circumstantial knowledge for making efficient decisions. The likelihood is the experiential knowledge about the fraction of times a feature observation has resulted in
a particular event. For example, clouds may cause it to rain and the likelihood is the
fraction of times observation of clouds resulted in rain. Circumstantial knowledge is
also a consideration in Bayesian formalism. It is called the prior. It is the background
information about the probability of the event to happen in a given situation. For