Artificial Intelligence Experiment #20
NextPreviousReset Experiment
Start Learning Hide Labels Hide Wires X-Ray Wires
Optic Nerve
Artificial Intelligence Introduction: Solving data bias

In this experiment we use a more diverse training data set in order to avoid data bias

This experiment is similar to the previous one. The only change we have made, is to include the white shapes on black background with the original training data of black shapes on white background. Start the learning process. Hint: Press on the Hide Wires button to make the learning process faster

Once learning is complete, drag the pattern slider under the letter. You will first see the black shapes on white background (the learning data) and then the same shapes, but white on a black background. Does our ANN recognize the white and black shapes?

The ANN has learned to recognise black shapes on a white background and white shapes on a black background. It can recognize the inverse (black and white swapped) shapes. Our network no longer has a data bias.

To avoid data bias, it is important to provide the ANN with a balanced training set, which is representatives of the actual data that the ANN will need to work with in real life.

The training of the ANN was significantly longer with the balanced learning dataset compared to the biased data set. Avoiding bias is hard and takes a lot of additional effort. That's because the information can be contradictory. Re-run this experiment again and also the one before. Write down the number of training cycles required for each experiment and build a ratio. How much slower was this experiment?

This text is displayed if your browser does not support HTML5 Canvas.
Training Cycles