The most popular and comprehensive Open Source ECM platform
Let us know more
Artificial Intelligence: Teaching Algorithms to Unlearn
It’s important to be able to unlearn. This can be true about bad personal habits, but we’re talking here about trained AI algorithms.
More frequently there are requirements for trained AI algorithm to unlearn specific data because the particular data is considered to be biased or because government regulations require data associated with citizens to not be used out of country.
Making AI unlearn is a very difficult requirement to fulfill. AI training takes time, is exceptionally resource intensive, and is derived in such a way that is generally too complex or virtually impossible to work the steps backwards to omit the influence and effect of a specific data point used in the original training.
Antonio Ginart, a PhD student at Stanford, wrote that “for many standard ML models, the only way to completely remove an individual’s data is to retrain the whole model from scratch on the remaining data, which is often not computationally practical.”
A group at the University of Pennsylvania under professor Aaron Roth is trying to solve the question: “Can we remove all influence of someone’s data when they ask to delete it, but avoid the full cost of retraining from scratch? As is common for a young field, there’s a gap between what this area aspires to do and what we know how to do now.” There is much work to be done, especially to find a method which might be generally applicable across different types of algorithms.
James Zou, an assistant professor of biomedical data science at Stanford University, told the Register that “deletion is difficult because most machine learning models are complex black boxes so it is not clear how a data point or a set of data point is really being used.”
A similar, but related question, is an attempt to quantify how much the accuracy an AI algorithm’s results declines as a result of removing data from the training set.
