Amartya Sanyal
@amartyasanyal.bsky.social
Assistant Professor @Dept. Of Computer Science, University of Copenhagen, Ex Postdoc @MPI-IS, ETHZ, PhD @University of Oxford, B.Tech @CSE,IITK.
Technically, we
1⃣ formalise the online learning unlearning (OLU) problem setting
2⃣ propose two styles of OLU algorithms
3⃣ In the Online Cvx Optimisation (OCO) framework, we nearly match the Regret guarantees of standard OCO without unlearning
1⃣ formalise the online learning unlearning (OLU) problem setting
2⃣ propose two styles of OLU algorithms
3⃣ In the Online Cvx Optimisation (OCO) framework, we nearly match the Regret guarantees of standard OCO without unlearning
May 15, 2025 at 3:47 PM
Technically, we
1⃣ formalise the online learning unlearning (OLU) problem setting
2⃣ propose two styles of OLU algorithms
3⃣ In the Online Cvx Optimisation (OCO) framework, we nearly match the Regret guarantees of standard OCO without unlearning
1⃣ formalise the online learning unlearning (OLU) problem setting
2⃣ propose two styles of OLU algorithms
3⃣ In the Online Cvx Optimisation (OCO) framework, we nearly match the Regret guarantees of standard OCO without unlearning
Reposted by Amartya Sanyal
#ICLR
»Differentially Private Steering for Large Language Model Alignment« by @anmolgoel.bsky.social, Yaxi Hu, Iryna Gurevych (@igurevych.bsky.social) & Amartya Sanyal (@amartyasanyal.bsky.social)
(2/🧵)
»Differentially Private Steering for Large Language Model Alignment« by @anmolgoel.bsky.social, Yaxi Hu, Iryna Gurevych (@igurevych.bsky.social) & Amartya Sanyal (@amartyasanyal.bsky.social)
(2/🧵)
January 27, 2025 at 11:03 AM
#ICLR
»Differentially Private Steering for Large Language Model Alignment« by @anmolgoel.bsky.social, Yaxi Hu, Iryna Gurevych (@igurevych.bsky.social) & Amartya Sanyal (@amartyasanyal.bsky.social)
(2/🧵)
»Differentially Private Steering for Large Language Model Alignment« by @anmolgoel.bsky.social, Yaxi Hu, Iryna Gurevych (@igurevych.bsky.social) & Amartya Sanyal (@amartyasanyal.bsky.social)
(2/🧵)
Thanks Christoph!
January 25, 2025 at 10:51 AM
Thanks Christoph!
@ccanonne.bsky.social : Steak holders
January 14, 2025 at 11:50 PM
@ccanonne.bsky.social : Steak holders
And I think a similar argument holds for synthetic data.
Synthetic data algorithms that don't provably account for privacy probably doesn't provide privacy.
But there are private synthetic data generation algorithms that do like @gautamkamath.com linked above.
Synthetic data algorithms that don't provably account for privacy probably doesn't provide privacy.
But there are private synthetic data generation algorithms that do like @gautamkamath.com linked above.
December 21, 2024 at 6:38 PM
And I think a similar argument holds for synthetic data.
Synthetic data algorithms that don't provably account for privacy probably doesn't provide privacy.
But there are private synthetic data generation algorithms that do like @gautamkamath.com linked above.
Synthetic data algorithms that don't provably account for privacy probably doesn't provide privacy.
But there are private synthetic data generation algorithms that do like @gautamkamath.com linked above.
Transformations like JL can indeed preserve privacy (arxiv.org/abs/1204.2136), while others may lead to (quantifiable) privacy degradation (arxiv.org/abs/2403.13041).
The point is perhaps that augmentations, by themselves, don’t inherently guarantee an increase or decrease in privacy.
The point is perhaps that augmentations, by themselves, don’t inherently guarantee an increase or decrease in privacy.
The Johnson-Lindenstrauss Transform Itself Preserves Differential Privacy
This paper proves that an "old dog", namely -- the classical Johnson-Lindenstrauss transform, "performs new tricks" -- it gives a novel way of preserving differential privacy. We show that if we take ...
arxiv.org
December 21, 2024 at 6:32 PM
Transformations like JL can indeed preserve privacy (arxiv.org/abs/1204.2136), while others may lead to (quantifiable) privacy degradation (arxiv.org/abs/2403.13041).
The point is perhaps that augmentations, by themselves, don’t inherently guarantee an increase or decrease in privacy.
The point is perhaps that augmentations, by themselves, don’t inherently guarantee an increase or decrease in privacy.
Reposted by Amartya Sanyal
Graduating with a PhD related to privacy and robustness in machine learning? Apply to this post-doc opening by @amartyasanyal.bsky.social: employment.ku.dk/faculty/?sho...
Postdoc in Privacy and Robustness of Machine Learning Algorithms
employment.ku.dk
December 3, 2024 at 9:57 AM
Graduating with a PhD related to privacy and robustness in machine learning? Apply to this post-doc opening by @amartyasanyal.bsky.social: employment.ku.dk/faculty/?sho...