Page 212 - DCAP208_Management Support Systems

Unit 12: Applications of Neural Network

Notes

Case Study Artificial Neural Network (ANN) Approach for an

Intelligent System

Artificial Neural Networks and Cognition
ANNs have massive parallelism, high connectivity, which tries to emulate the Biological
neurons and their synapses. NNs are especially useful for classification and function
approximation/mapping problems. The networks use Input/Output layers and none or
few hidden layers (HLs). Every hidden and output unit has its own bias term Learning
algorithms can be supervised or unsupervised. In supervised learning, the correct results
(desired outputs) are known and are given to the NN during training so that the NN can
adjust its weights to try match its outputs to the target values. After training, NN is tested
by giving only input values, to see how closely it predicts target values. LM, Momentum
Step are common learning algorithms. Performances of different network topologies can
be compared by evaluating the error function. Various networks are trained by
minimization of an appropriate error function defined with respect to a training data set.
Performance of networks are compared by evaluating the error function using an
independent cross-validation (CV) set.
Network having smallest error with respect to CV set is selected (hold out method).
Performance of the selected network is confirmed using a third independent set of data
called a test set to avoid overfitting.
Multilayer Perceptron (MLP)
MLP has one or more HLs with a linear combination function is the inner product of
inputs, weights and a bias. Activation function is logistic or tanh function. Inputs are fully
connected to the first HL, each HL is fully connected to the next, and the last HL is fully
connected to the outputs. MLPs use supervised learning or backpropagation. Designing
and training an MLP requires (i) selecting number of HLs, (ii) number of neurons to be
used in each HL, (iii) avoiding local minima, (iv) converging to an optimal solution in a
reasonable period of time, (v) validating NN to test for overfitting. MLPs are used in
classification problems, in fitness approximations.

Linear-Regression
Linear regression assumes that expected value of the output given an input, is linear.
BAYES rule is used to obtain posterior distribution for given data. Maximum likelihood
estimation helps in predictions.
Tonic-Shift-Property
CCM is a highly scientific and evolved form of classical music which uses a heptatonic
scale of seven notes symbolically given as S, R, G, M, P, D, N in ascending order of
frequencies. It is not equi-tempered scale like the Western, since notes have varying ratios
with their successive notes. Two notes S and P are inherently stable with no variants. The
remaining 5 notes R, G, M, D, N have variants. M has 2 variants -M1 and M2. They result in
12 semi-tones as shown in TABLE 1. In CCM a full scale (heptatonic) consists of all the 7
notes in both ascending and descending called a ‘melakartha’ (‘Janaka’) raga if i) all the 7
notes in ascending and descending are identical.
Contd...

LOVELY PROFESSIONAL UNIVERSITY 205