Training Procedure
A specific subset of a collected data set
is used for training purposes and the remainder of the data set is used
for testing purposes. This permits an immediate understanding of the
efficiency of the Neural Network during Neural Network training.
The training graph screen (shown above) presents two
graphs.
- The top graph indicates the Neural Network response
to the training data set.
- The bottom graph indicates the Neural Network
response to the test data set.
In this example, the Neural Network is being trained to
identify a specific object - in this case, a container of water.
- When the Neural Network is being trained with the
specific object, it is instructed to provide an output value of 0.9
units.
- When the Neural Network is being trained with no
specific object, it is instructed to provide an output value of 0.1
units.
Training the Neural Network
At the start of training, the Neural Network can not
determine the difference between either of the two conditions and it provides
an output value of around 0.5 units. However, as the Neural Network
is stepped through its training algorithm, it starts to recognize a
pattern between the two waveform data sets, and the resultant Neural
Network output result for each condition separates. The training
output data sets have been placed in the top graph and it can be seen that
as the Neural Network trains that it begins to correctly recognize which data
set belongs to each respective condition (for example, the specific object
data is being correctly trained to 0.9 units).
Testing the Neural Network
What is really interesting is that the lower graph also
indicates that the Neural Network can recognize a difference between the
two testing datasets - the test datasets are data which the Neural Network
has never seen before.
Future Enhancements
In order to make the system more accurate, a new data
collection interface is in the process of being designed.
This is an ongoing research project...
