Melis İlayda Bal
@melisilaydabal.bsky.social
PhD student @Max Planck Institute for Intelligent Systems | working on optimization & game-theoretic frameworks for robust and efficient ML | organizer @twiml.bsky.social | https://melisilaydabal.github.io
📄Preprint: arxiv.org/abs/2502.17121
Many thanks to my co-authors @cevherlions.bsky.social and Michael Muehlebach!
Looking forward to presenting this at ICLR 2025! If you're interested in adversarial robustness, I’d be happy to connect!
Many thanks to my co-authors @cevherlions.bsky.social and Michael Muehlebach!
Looking forward to presenting this at ICLR 2025! If you're interested in adversarial robustness, I’d be happy to connect!
Adversarial Training for Defense Against Label Poisoning Attacks
As machine learning models grow in complexity and increasingly rely on publicly sourced data, such as the human-annotated labels used in training large language models, they become more vulnerable to ...
arxiv.org
February 28, 2025 at 10:01 PM
📄Preprint: arxiv.org/abs/2502.17121
Many thanks to my co-authors @cevherlions.bsky.social and Michael Muehlebach!
Looking forward to presenting this at ICLR 2025! If you're interested in adversarial robustness, I’d be happy to connect!
Many thanks to my co-authors @cevherlions.bsky.social and Michael Muehlebach!
Looking forward to presenting this at ICLR 2025! If you're interested in adversarial robustness, I’d be happy to connect!
...an adversarial training framework designed to enhance robustness against these attacks.
🔹Defense formulated as a bilevel optimization framework using kernel SVMs.
🔹Adapts against poisoned labels, improving robust accuracy.
🔹Scalable and outperforms robust baselines under strong attacks.
🔹Defense formulated as a bilevel optimization framework using kernel SVMs.
🔹Adapts against poisoned labels, improving robust accuracy.
🔹Scalable and outperforms robust baselines under strong attacks.
February 28, 2025 at 9:58 PM
...an adversarial training framework designed to enhance robustness against these attacks.
🔹Defense formulated as a bilevel optimization framework using kernel SVMs.
🔹Adapts against poisoned labels, improving robust accuracy.
🔹Scalable and outperforms robust baselines under strong attacks.
🔹Defense formulated as a bilevel optimization framework using kernel SVMs.
🔹Adapts against poisoned labels, improving robust accuracy.
🔹Scalable and outperforms robust baselines under strong attacks.