Learning to Unlearn: Why Data Scientists and AI Practitioners Should Understand Machine Unlearning
Explore the intersections between privacy and AI with a guide to removing the impact of individual data points in AI training using the SISA technique applied to Convolutional Neural Networks (CNNs) using Python.
To the date that this article is being written and based on World Bank data, over 32% of the world's population (approximately 8 billion) is under twenty years old. This means that approximately 2.6 billion people were born in the social media era, and it's highly probable that almost all their lives have been registered online, by their parents, their inner circle, or in the end by themselves (depending on their attachment to social media as well as their network). If we add the people who are between their twenties and fifties, we have an extra 3.3 billion people who, to some extent, have a part of their lives registered online in different sources and formats (images, comments, videos, etc.). Of course, we can adjust the numbers considering the people over fifty, or that not everyone in the world has access to or uses the internet (at least more than 35% don't have access or use it, based on World Bank estimations in 2021), but I'm sure you understand my point. There is a significant amount of our lives registered in today's digital world.
Another high probability or maybe certainty (we could ask again OpenAI's CTO
