Explained what learned models predict

This essay was originally created for a university application. But before the essay gathers dust on my hard drive, Iā€™d rather share it here šŸ˜„.

Abstract

There are many problems in computer science that cannot or can only be solved with extreme difficulty using pre-programmed rules. An example here would be the recognition and classification of images. Here, machine learning (ML) algorithms offer a good solution approach by recognizing regularities from previous examples, storing them, and applying them to the new images. However, the safety and reliability of machine-learning systems cannot be readily assessed because the individual steps in the learning process of an ML system are not easily comprehensible to a human, making the decision path very opaque. Given the ever-increasing impact of ML systems in our lives, it is very important that these systems are reliable since wrong or unintended results could have extreme consequences.

This paper will discuss a small sample of possible causes of reliability problems and present some solutions.

Link to the document