Making ML model error an information source - Marta Hendler - code::dive 2023
Making ML model error an information source
In recent years, methods of artificial intelligence, and in particular machine learning algorithms, have settled in the technology industry for good and gained immense popularity. They have been successfully applied to a variety of problems in numerous technological areas, one of which is the mobile network industry. However, due to the growing popularity and nearly limitless potential applications of these methods, machine learning experiments are easily reduced to rigidly defined phases, and the applications of the emerging models are limited to their predictive capabilities.
It is worth noting that the error produced by the model – the key concern when employing ML methods – can be used not only to assess model’s accuracy and overall performance but also to derive useful information about the state of the system which is observed during the experiment. To support this statement, we provide an example of a ML-based method for the assessment of configuration changes and their impacts on performance indicators in mobile networks that employs such approach. We present how to define the measures to be able to draw conclusions about the gain or loss in network performance based on the analysis of residuals of machine learning model.
About the speaker - Marta Hendler
Marta has a two year experience of working at Nokia, where she currently works as a Data Scientist.
She is also a PhD student in biomedical engineering, where she deals with data related to brain hemodynamics.
Apart from work and doctoral studies, Marta enjoys travelling, practicing yoga and brewing alternative coffee. She also reads all kinds of books - from reportage and classics of literature to fantasy sagas for teenagers.
Смотрите видео Making ML model error an information source - Marta Hendler - code::dive 2023 онлайн без регистрации, длительностью часов минут секунд в хорошем качестве. Это видео добавил пользователь code::dive conference 01 Январь 1970, не забудьте поделиться им ссылкой с друзьями и знакомыми, на нашем сайте его посмотрели 132 раз и оно понравилось 1 людям.