Publications | Michael Aernihttps://www.michaelaerni.com/publication/PublicationsWowchemy (https://wowchemy.com)en-ushttps://www.michaelaerni.com/media/icon_hua9c6b47fafe860101b75091fc8397190_346273_512x512_fill_lanczos_center_3.pngPublicationshttps://www.michaelaerni.com/publication/Modal Aphasia: Can Unified Multimodal Models Describe Images From Memory?https://www.michaelaerni.com/publication/modal-aphasia/Wed, 22 Oct 2025 00:00:00 +0000https://www.michaelaerni.com/publication/modal-aphasia/Measuring Non-Adversarial Reproduction of Training Data in Large Language Modelshttps://www.michaelaerni.com/publication/measuring-non-adversarial-reproduction-of-training-data-in-large-language-models/Tue, 11 Feb 2025 00:00:00 +0000https://www.michaelaerni.com/publication/measuring-non-adversarial-reproduction-of-training-data-in-large-language-models/Evaluations of Machine Learning Privacy Defenses are Misleadinghttps://www.michaelaerni.com/publication/evaluations-of-machine-learning-privacy-defenses-are-misleading/Mon, 14 Oct 2024 00:00:00 +0000https://www.michaelaerni.com/publication/evaluations-of-machine-learning-privacy-defenses-are-misleading/Strong inductive biases provably prevent harmless interpolationhttps://www.michaelaerni.com/publication/strong-inductive-biases-provably-prevent-harmless-interpolation/Wed, 21 Sep 2022 00:00:00 +0000https://www.michaelaerni.com/publication/strong-inductive-biases-provably-prevent-harmless-interpolation/Interpolation can hurt robust generalization even when there is no noisehttps://www.michaelaerni.com/publication/interpolation-can-hurt-robust-generalization-even-when-there-is-no-noise/Mon, 09 Aug 2021 00:00:00 +0000https://www.michaelaerni.com/publication/interpolation-can-hurt-robust-generalization-even-when-there-is-no-noise/<p>We presented two related workshop papers:</p> <ul> <li><a href="https://www.michaelaerni.com/publication/surprising-benefits-of-ridge-regularization-for-noiseless-regression/">Surprising benefits of ridge regularization for noiseless regression</a> in ICML 2021 Workshop on Overparameterization: Pitfalls and Opportunities</li> <li><a href="https://www.michaelaerni.com/publication/maximizing-the-robust-margin-provably-overfits-on-noiseless-data/">Maximizing the robust margin provably overfits on noiseless data</a> in ICML 2021 Workshop on Adversarial Machine Learning</li> </ul>Surprising benefits of ridge regularization for noiseless regressionhttps://www.michaelaerni.com/publication/surprising-benefits-of-ridge-regularization-for-noiseless-regression/Sat, 24 Jul 2021 16:00:00 +0000https://www.michaelaerni.com/publication/surprising-benefits-of-ridge-regularization-for-noiseless-regression/Maximizing the robust margin provably overfits on noiseless datahttps://www.michaelaerni.com/publication/maximizing-the-robust-margin-provably-overfits-on-noiseless-data/Sat, 24 Jul 2021 11:45:00 +0000https://www.michaelaerni.com/publication/maximizing-the-robust-margin-provably-overfits-on-noiseless-data/