Further Experiments in Teaching Neural Networks

The basis of this idea was inspired by some work that can be found in the paper Varley et al (2005). The idea is to use spreadsheets as a tool for teaching neural networks. These approaches were developed for the teaching of these approaches and the videos were recorded in a class whilst teaching the concept. Both Excel and Google Sheets have been tried in all three of the approaches.

1. Creating a single neuron
The idea was to use a spreadsheet to replicate the neuron. The idea was to get the students to build the neuron step by step, starting with the inputs, then the weights, then the weighted sum and threshold for a simple neuron. Personally, I like the approach for two reasons; firstly the concept that only a single set of weights is used is sneaked in, the second is builds all the stages up gradually and visually.
Video below shows the building of the model and it's use. 



2. Training a single neuron.
This approach extends the ideas from the approach above, quickly building a single neuron, but expands into training a neuron. Cutting and pasting the blocks to show the idea of epochs.

Training is via the delta rule (change in the weight [x] = learning coefficient * input [x] * (what output we wanted - actual output from the neuron). I like the approach because it seems to show the weights being arrived at in a very mechanical way by repeating actions -  machine learning is not magic!

The video below shows the stages.



3. Building a simple Neural Network
Previously only a single neuron was produced. In this approach, the ideas from build a neuron in Excel activity are extended to three neurons connected in the same worksheet. The exercise then connects the outputs from two neurons (which have a common input) as the inputs of a third neuron building an XOR gate (which a single simple neuron can not implement). The video below shows all of the stages.


 




4. Where next?
It would be nice to extend the idea further to have the training of a simple neural network above. If anyone manages this please add the link into the comments.


Bibliography
VARLEY, M; PEAK, M; HEYS, J; COLLINS, G; KONSTANTARAS A,  VALLIANATOS, F; PICTON P (2005) Spreadsheet Software as a Teaching Tool for Concepts in Electronic Engineering, [Online] http://www.wseas.us/e-library/conferences/2005athens/ee/papers/507-162.pdf accessed on: 20/2/2016

Turner, S. J. (2017) Experience of using spreadsheets as a bridge in the understanding of AI techniques. Paper presented to: 13th China Europe Symposium on Software Engineering Education (CEISEE), Athens, Greece, 24-25 May 2017.


All views and opinions are the author's and do not necessarily reflected those of any organisation they are associated with

All views and opinions are the author's and do not necessarily reflected those of any organisation they are associated with. Twitter: @scottturneruon

Excel to build and training a single artificial neurone

Two videos produced within the Department of Computing, University of Northampton, Uk on using Microsoft Excel to simulate a single artificial neurone, as a step towards building a neural network.

In the first video, Excel is used to build and show the principles of a single neurone.



In the second video the ideas from the video above are now extended to show, in a relatively simple way, how a single neurone can be trained.



What would be interesting to find a way to build this idea into a simple neural network. Comments and ideas are very welcome.


If you'd like to find out more about Computing at the University of Northampton go to: www.computing.northampton.ac.uk. All views and opinions are the author's and do not necessarily reflected those of any organisation they are associated with

update: Review of Artificial Neural Networks (ANN) applied to corrosion monitoring

S Mabbutt, P Picton, P Shaw and S Black (2012) Review of Artificial Neural Networks (ANN) applied to corrosion monitoring J. Phys.: Conf. Ser. 364 012114 doi:10.1088/1742-6596/364/1/012114 

To read the paper go to: http://iopscience.iop.org/1742-6596/364/1/012114/pdf/1742-6596_364_1_012114.pdf



Abstract:The assessment of corrosion within an engineering system often forms an important aspect of condition monitoring but it is a parameter that is inherently difficult to measure and predict. The electrochemical nature of the corrosion process allows precise measurements to be made. Advances in instruments, techniques and software have resulted in devices that can gather data and perform various analysis routines that provide parameters to identify corrosion type and corrosion rate. Although corrosion rates are important they are only useful where general or uniform corrosion dominates. However, pitting, inter-granular corrosion and environmentally assisted cracking (stress corrosion) are examples of corrosion mechanisms that can be dangerous and virtually invisible to the naked eye. Electrochemical noise (EN) monitoring is a very useful technique for detecting these types of corrosion and it is the only non-invasive electrochemical corrosion monitoring technique commonly available. Modern instrumentation is extremely sensitive to changes in the system and new experimental configurations for gathering EN data have been proven. In this paper the identification of localised corrosion by different data analysis routines has been reviewed. In particular the application of Artificial Neural Network (ANN) analysis to corrosion data is of key interest. In most instances data needs to be used with conventional theory to obtain meaningful information and relies on expert interpretation. Recently work has been carried out using artificial neural networks to investigate various types of corrosion data in attempts to predict corrosion behaviour with some success. This work aims to extend this earlier work to identify reliable electrochemical indicators of localised corrosion onset and propagation stages.

To read the paper go to: http://iopscience.iop.org/1742-6596/364/1/012114/pdf/1742-6596_364_1_012114.pdf


Related postings:
PREDICTING LEATHER HANDLE LIKE AN EXPERT BY ARTIFICIAL NEURAL NETWORKS
Subjective Measurement of Leather Handle

Artificial Neural Networks (ANN) and corrosion monitoring

A paper has recently being published by members of the department of engineering, University of Northampton reviewing neural networks applied to corrosion conditioning monitoring.


Abstract
The assessment of corrosion within an engineering system often forms an important aspect of condition monitoring but it is a parameter that is inherently difficult to measure and predict. The electrochemical nature of the corrosion process allows precise measurements to be made. Advances in instruments, techniques and software have resulted in devices that can gather data and perform various analysis routines that provide parameters to identify corrosion type and corrosion rate. Although corrosion rates are important they are only useful where general or uniform corrosion dominates. However, pitting, inter-granular corrosion and environmentally assisted cracking (stress corrosion) are examples of corrosion mechanisms that can be dangerous and virtually invisible to the naked eye. Electrochemical noise (EN) monitoring is a very useful technique for detecting these types of corrosion and it is the only non-invasive electrochemical corrosion monitoring technique commonly available. Modern instrumentation is extremely sensitive to changes in the system and new experimental configurations for gathering EN data have been proven. In this paper the identification of localised corrosion by different data analysis routines has been reviewed. In particular the application of Artificial Neural Network (ANN) analysis to corrosion data is of key interest. In most instances data needs to be used with conventional theory to obtain meaningful information and relies on expert interpretation. Recently work has been carried out using artificial neural networks to investigate various types of corrosion data in attempts to predict corrosion behaviour with some success. This work aims to extend this earlier work to identify reliable electrochemical indicators of localised corrosion onset and propagation stages.


Reference
Review of Artificial Neural Networks (ANN) applied to corrosion monitoring

S Mabbutt et al 2012 J. Phys.: Conf. Ser. 364 012114

doi:10.1088/1742-6596/364/1/012114