Conclusion

In this article, we have seen two examples of the application of learning methods. The first was a classification problem, concerning the prediction of air traffic controllers' workload, with three categories: low, normal or excessive. The second was a regression problem, in which we sought to predict the future altitude of an aircraft, or certain missing parameters in the physical model used to calculate this altitude. In both cases, we were able to see the practical benefits of learning methods, in terms of improving the quality of predictive models.

In these and other applications, it is beneficial (and even advisable) to draw on theoretical considerations to make judicious modeling choices. For example, the choice to minimize a cross-entropy and use a softmax function in the hidden layer of our neural network in paragraph