iiinova — where your digital self keeps evolving

Aviraj (Avi) Newatia

@projectavi.bsky.social

Machine Learning Researcher. Undergraduate Computer Scientist. University of Toronto & Vector Institute.

Aviraj (Avi) Newatia @projectavi.bsky.social · 14/12/2024

6/6 In conclusion, RELOAD is an effective algorithm for unlearning arbitrary parts of the training set, and provides strong privacy guarantees for forgotten data.

000

Aviraj (Avi) Newatia @projectavi.bsky.social · 14/12/2024

5/6 Using TabNet attention masks we show how RELOAD removes dependence of model inference on forgotten features.

000

Aviraj (Avi) Newatia @projectavi.bsky.social · 14/12/2024

4/6 We conducted experiments on forgetting random samples and entire features from the training set, consistently outperforming unlearning baselines and protecting user privacy.

000

Aviraj (Avi) Newatia @projectavi.bsky.social · 14/12/2024

3/6 Key Idea: We compare cached end-of-training gradients to those on the remaining data to identify parameters in the model to reset.

000

Aviraj (Avi) Newatia @projectavi.bsky.social · 14/12/2024

2/6 Key Motivation: In unlearning, we typically require access to the set of data being forgotten. How can we unlearn, without that data?

000

Aviraj (Avi) Newatia @projectavi.bsky.social · 14/12/2024

1/6 Presenting "Unlearning Tabular Data without a 'Forget Set'"! We explore a new unlearning algorithm RELOAD in tabular learning. Drop by @neuripsconf.bsky.social Workshop on Table Representation Learning (@trl-research.bsky.social): - SAT 14 Dec from 2:30pm-3:15pm! - East Meeting Room 11-12

511