Have a UROP opening you would like to submit?
Please fill out the form.
Selective Forgetting in Deep Neural Networks
MAS: Media Arts and Sciences
20 October 2020
Ayush Chopra, firstname.lastname@example.org
Context: Users voluntarily provide lot of personal data to online services, such as Facebook, Google, and Amazon, in exchange for services. There is growing regulatory focus (w/ GDPR) that users should be able to revoke access to their data if they no longer find the exchange of data for services worthwhile. Typically, this user data does not sit in databases, but is used to build predictive models such as deep neural networks. We intend to explore methods for selectively forgetting of a particular subset of the data used for training a deep neural network, without retraining from scratch. Setup: Input is a pre-trained model, say image classifier, and a subset of the train data which needs to be forgotten (called forget dataset). The task is to fine-tune the model so as to minimize performance on this forget data while also preserving performance on the rest of the training data (called retained dataset) Objective: Submit conference paper at a top tier ML/CV conference.
Required: Background in Programming, Math, Machine Learning Preferred: Background in Deep Learning, Computer Vision/NLP