4.1 Minimising errors (simple)
You can train the neural network manually by changing the neuron’s weights and threshold value with the plus and minus signs. Each change results in the current error of the neural network being displayed in the adjacent box. This allows you to see whether the changes have already led in the right direction or not. Remember: The error of the network is the difference, roughly speaking, between the desired and the actual output.
The more complicated a neural network gets, the more challenging this task becomes, of course – the same holds true for the applied learning algorithm if you simply let the network train itself by clicking on Train. This is because even a minimal change to just one weight has an immediate impact on all subsequent neurons in the layer behind it and ultimately also on the output of the network. The algorithm therefore needs to optimise all weights in an optimally coordinated manner. With somewhat more complicated mathematics, which goes a little beyond the school level, this can however be solved quite well algorithmically.
Even if you cannot manually click through the exact strategy of the real learning algorithm, going through this and the slightly more complex example at the end of the chapter will give you a sense of how a global optimisation algorithm must work in principle to optimally adjust all weights and thresholds simultaneously. In doing so, you can use a trick of the real learning algorithm by changing, for instance, the weights (or the threshold) in reverse order, starting from the back.
Instructions
- Use the checkboxes to select which training data should be loaded.
- Click
Newto select the pre-selected values for a simple separation, a new randomly selected Boolean function or new random numbers as a data set. - Now click on the plus or minus signs in the figure to change the weights or the threshold value. The current error of the network is automatically displayed. If the error is small enough, you will receive a pop-up message with your required time.
- Click
Resetto undo all your current changes to the weights and threshold. - Click
Trainto start or resume training. - In the figure on the left, blue stands for negative values and red stands for positive values.
Task
Try out many different examples and minimise the network’s error for each one until you get a sense of how to adjust the threshold and weights to quickly minimise the error for any given data set.
Share this page
