Communications of the ACM
New Technique Wipes Out nwanted Data
Credit: Moreco
Researchers at Lehigh and Columbia University have developed a machine-learning method that involves making such systems forget the data's "lineage" so they can remove the data and undo its effects and allow future operations to run as if the data never existed. Although the concept of "machine unlearning" is well-established, the researchers have developed a way to do it faster and more effectively than can be done using current methods. Effective machine-unlearning techniques can help improve the privacy and security of raw data.
The new machine-unlearning method is based on the fact that most learning systems can be converted into a form that can be updated incrementally without costly retraining from scratch. The approach introduces a layer of a small number of summations between the learning algorithm and the training data to eliminate dependency on each other; the learning algorithms depend only on the summations and not on individual data. The method enables machine-learning systems to unlearn a piece of data and its lineage without rebuilding the models and features that predict relationships between pieces of data. Recomputing a small number of summations would remove the data and its lineage completely, and is much faster than by retraining the system.
From Lehigh University
View Full Article
Abstracts Copyright © 2016 Information Inc., Bethesda, Maryland, USA
No entries found