<< Chapter < Page Chapter >> Page >

Neural network implementation

For our classification task, we implement a multiclass artificial neural network. Artificial neural networks are computing structures that attempt to mimic the brain. They perform by means of distributed processing among nodes, which correspond to neurons. Connection strength between individual neurons collectively determines the network’s activity, similar to biological synapse performance. Due to the computational flexibility inherent in these large, distributed systems, neural networks excel at nonlinear tasks such as facial or speech recognition. We expected that a neural network would excel at our task: emotion classification.

Network specifications

Example of a multilayer perceptron.

We utilize the common multilayer perceptron neural network with neuron activation modeled by a sigmoid.

Neuron model

Courtesy of coursera.org

Neurons are heavily connected – one connects to many, and many to others. While each neuron takes in multiple inputs, each input is not weighted equally. The weight determines how strongly an input will affect a neuron’s output, hθ. The activation function Fa determines the relation between the inputs and the output.

The neurons of our network are characterized by the sigmoid activation function.

The sigmoid neuron has two distinct advantages. First, the sigmoid activation function is differentiable. This simplifies the function used in calculating cost for training. Second, the sigmoid activation function retains thresholding – output tends toward either 1 or 0. This enables the neurons to behave as ‘on/off’ nonlinear switches.

Network architecture

We use the same type of network as displayed in fig. 1: a multilayer perceptron. This network is feed forward, fully connected, and contains>= three layers. Feed forward means that signals travel straight through the network, from input to output, via hidden layers. No connections exist between neurons of the same layer. Fully connected means that each neuron of a previous layer has a connection between every neuron of the next layer. Feed forward systems have comparatively manageable training mechanisms, and so are preferred to other connection systems.

The input layer of a multiperceptron takes in one feature per node. The output layer contains one class per node. This layer yields the probability that an input is that particular class. The hidden layers perform operations on the inputs to yield the output. The hidden layers are thought to flesh out the distinct features that relate input to output, though what they flesh out is difficult to obtain as well as to understand.

For our application, speech features such as formant locations and power statistics are fed into 14 input nodes, and the emotion category is extracted from the 15 output nodes. The number of input nodes depends on the number of speech features, 14, that we are classifying by. The number of output nodes corresponds to the 15 emotion classes samples are categorized into. In between this input and output layer lies one hidden layer of approximately 100 artificial neurons. Our network necessitated such a large number of neurons due to the complex relation between speech features and emotion.


To tailor the network for our classification task, we train its weights via feedforward-back propagation, a popular means to minimize a neural network’s cost function: J(Θ). The process is as follows:

  • Randomly initialize weights between nodes. We take Θ = weight matrix.
  • Implement forward propagation to determine J(Θ).
Courtesy of coursera.org

In forward propagation, all training samples are fed into the initialized neural network to obtain estimated outputs, which are used to determine J(Θ). To prevent overfitting, the cost function includes a regularization term with λ = 1, which enables control over the overall magnitudes of Θ parameters.

  • Implement back propagation to determine the gradient of J(Θ). In back propagation, the error derivatives of latter layer weights are used to calculate error derivatives of former layer weights, up until all weight derivatives are found.
  • Use an optimization algorithm (in our case, batch gradient descent) to minimize J(Θ), and obtain a the correct set of weights. To increase the speed of gradient descent, implement feature scaling, wherein each input feature is first shifted by the mean, and then scaled by the standard deviation.

Neural networks tend to have multiple local optima. To determine the global optima, the back propagation algorithm is run through multiple iterations (we use 1000).

Performance testing

We evaluate an independent test database with the trained neural network to test the network’s performance.

Useful references

To learn more about neural networks, check out Andrew Ng’s Machine Learning or Geoffrey Hinton’s Neural Networks for Machine learning, as found on coursera.org. Our neural network code is modified from Andrew Ng’s course code.

Questions & Answers

Is there any normative that regulates the use of silver nanoparticles?
Damian Reply
what king of growth are you checking .?
What fields keep nano created devices from performing or assimulating ? Magnetic fields ? Are do they assimilate ?
Stoney Reply
why we need to study biomolecules, molecular biology in nanotechnology?
Adin Reply
yes I'm doing my masters in nanotechnology, we are being studying all these domains as well..
what school?
biomolecules are e building blocks of every organics and inorganic materials.
anyone know any internet site where one can find nanotechnology papers?
Damian Reply
sciencedirect big data base
Introduction about quantum dots in nanotechnology
Praveena Reply
what does nano mean?
Anassong Reply
nano basically means 10^(-9). nanometer is a unit to measure length.
do you think it's worthwhile in the long term to study the effects and possibilities of nanotechnology on viral treatment?
Damian Reply
absolutely yes
how to know photocatalytic properties of tio2 nanoparticles...what to do now
Akash Reply
it is a goid question and i want to know the answer as well
characteristics of micro business
for teaching engĺish at school how nano technology help us
Do somebody tell me a best nano engineering book for beginners?
s. Reply
there is no specific books for beginners but there is book called principle of nanotechnology
what is fullerene does it is used to make bukky balls
Devang Reply
are you nano engineer ?
fullerene is a bucky ball aka Carbon 60 molecule. It was name by the architect Fuller. He design the geodesic dome. it resembles a soccer ball.
what is the actual application of fullerenes nowadays?
That is a great question Damian. best way to answer that question is to Google it. there are hundreds of applications for buck minister fullerenes, from medical to aerospace. you can also find plenty of research papers that will give you great detail on the potential applications of fullerenes.
what is the Synthesis, properties,and applications of carbon nano chemistry
Abhijith Reply
Mostly, they use nano carbon for electronics and for materials to be strengthened.
is Bucky paper clear?
carbon nanotubes has various application in fuel cells membrane, current research on cancer drug,and in electronics MEMS and NEMS etc
so some one know about replacing silicon atom with phosphorous in semiconductors device?
s. Reply
Yeah, it is a pain to say the least. You basically have to heat the substarte up to around 1000 degrees celcius then pass phosphene gas over top of it, which is explosive and toxic by the way, under very low pressure.
Do you know which machine is used to that process?
how to fabricate graphene ink ?
for screen printed electrodes ?
What is lattice structure?
s. Reply
of graphene you mean?
or in general
in general
Graphene has a hexagonal structure
On having this app for quite a bit time, Haven't realised there's a chat room in it.
what is biological synthesis of nanoparticles
Sanket Reply
how did you get the value of 2000N.What calculations are needed to arrive at it
Smarajit Reply
Privacy Information Security Software Version 1.1a
Got questions? Join the online conversation and get instant answers!
Jobilize.com Reply

Get the best Algebra and trigonometry course in your pocket!

Source:  OpenStax, Robust classification of highly-specific emotion in human speech. OpenStax CNX. Dec 14, 2012 Download for free at http://cnx.org/content/col11465/1.1
Google Play and the Google Play logo are trademarks of Google Inc.

Notification Switch

Would you like to follow the 'Robust classification of highly-specific emotion in human speech' conversation and receive update notifications?