Interpolation Learning With Minimum Description Length

We prove that the Minimum Description Length learning rule exhibits tempered overfitting. We obtain tempered agnostic finite sample learning guarantees and characterize the asymptotic behavior in the presence of random label noise.

cs.LG cs.DS cs.IT stat.ML Neural Networks and Applications Fuzzy Logic and Control Systems Rough Sets and Fuzzy Logic

View graph Source

Paper

Similar papers

Minimum Description Length Recurrent Neural NetworksNur Lan, Michal Geyer, Emmanuel Chemla et al.2022 · Transactions of the Association for Computational Linguistics In Library

Minimum Description Length and Generalization Guarantees for Representation LearningMilad Sefidgaran, Abdellatif Zaidi, Piotr Krasnowski2024 · arXiv (Cornell University) · 16 citations In Library

Sequential Learning Of Neural Networks for Prequential MDLJörg Bornschein, Yazhe Li, Marcus Hütter2022 · arXiv (Cornell University) · 16 citations In Library

Minimum Description Length ControlMoskovitz, T, Kao, C, Sahani, M et al.2023 · UCL Discovery (University College London) In Library

Minimum Description Length ControlMoskovitz, Ted, Kao, Ta-Chu, Sahani, Maneesh et al.2022 · arXiv (Cornell University) · 1 citation In Library

Prediction Under Imperfect Compression: A Theory of Approximate MDLQian Li, Xinyu Mao, Shang-Hua Teng et al.2026 · 0 citations In Library

Increasing generalizability via the principle of minimum description lengthWes Bonifay2022 · Behavioral and Brain Sciences In Library

Minimum Description Length Recurrent Neural NetworksNur Lan, Michal Geyer, Emmanuel Chemla et al.2021 · arXiv (Cornell University) · 19 citations In Library

Bridging the Empirical-Theoretical Gap in Neural Network Formal Language Learning Using Minimum Description LengthNur Lan, Emmanuel Chemla, Roni Katzir2024 In Library

Approaching explanatory adequacy in phonology using Minimum Description LengthEzer Rasin, Iddo Berger, Nur Lan et al.2021 · DOAJ (DOAJ: Directory of Open Access Journals) In Library

Patents near this paper

MULTILABEL LEARNING WITH LABEL RELATIONSHIPSUS 11,769,087 · MICROSOFT TECHNOLOGY LICENSING, LLC · 2023 METHOD AND APPARATUS FOR MACHINE LEARNINGUS 8,972,307 · GOOGLE INC. · 2015 LEVERAGING UNLABELED DATA WITH A PROBABILISTIC GRAPHICAL MODELUS 7,937,264 · MICROSOFT CORPORATION · 2011 FASTER MINIMUM ERROR RATE TRAINING FOR WEIGHTED LINEAR MODELSUS 9,098,812 · MICROSOFT CORPORATION · 2015 INCREMENTAL LEARNER VIA AN ADAPTIVE MIXTURE OF WEAK LEARNERS DISTRIBUTED ON A NON-RIGID BINARY TREEUS 10,262,274 · ASELSAN ELEKTRONIK SANAYI VE TICARET ANONIM SIRKETI · 2019

Source of record

arXiv