Page 192 - DCAP208_Management Support Systems

Unit 11: Neural Networks

Notes

Notes The learning algorithm being used determines how the neural interconnection
weights are corrected due to differences in the actual and desired output for a member of
the training set.

Updating of the network’s interconnection weights continues until the stopping criteria of the
training algorithm are met (e.g., all cases must be correctly classified within a certain tolerance
level). Alternatively, in unsupervised learning, there are no target answers that the network
tries to learn. Instead, the neural network learns a pattern through repeated exposure. Thus, this
kind of learning can be envisioned as the neural network appropriately self-organizing or
clustering its neurons related to the specific desired task. Multilayer, feedforward neural networks
are a class of models that show promise in classification and forecasting problems. As the name
implies, these models structurally consist of multiple layers of neurons. Information is passed
through the network in one direction, from the input layers of the network, through one or
more hidden layers, toward the output layer of neurons. Neurons of each layer are connected
only to the neurons of the subsequent layer.

11.1.6 Learning in ANN

An important consideration in an ANN is the use of an appropriate learning algorithm (or
training algorithm). Learning algorithms specify the process by which a neural network learns
the underlying relationship between input and outputs, or just among the inputs. There are
hundreds of them. Learning algorithms in ANN can also be classified as supervised learning
and unsupervised learning (see Figure 11.8).
Figure 11.8: Taxonomy of ANN Architectures and Learning Algorithms

Source: http://www70.homepage.villanova.edu/matthew.liberatore/Mgt2206/turban_online_ch06.pdf
Supervised learning uses a set of inputs for which the appropriate (i.e., desired) outputs are
known.

LOVELY PROFESSIONAL UNIVERSITY 185