arXiv:2608. 03386v1 Announce Type: new Abstract: Contemporary deep learning methods generalize well even when they fit their training data perfectly, a phenomenon known as benign interpolation.
By Tom F. Sterkenburg, Daniel A. Herrmann, Jan-Willem Romeijn
The paper titled "Pessimistic Meta-Induction and Its Limits: Lessons from Frequentist Statistics and Machine Learning Theory" critiques the pessimistic meta-inductive argument against scientific realism by attacking its inductive step rather than its historical premise. It introduces a new challenge, drawing on frequentist statistics, machine learning, and formal epistemology to assess induction through convergence to truth. The authors argue that ordinary enumerative induction can achieve convergence everywhere, whereas meta-induction fails to achieve even almost everywhere convergence, and in contexts where meta-induction applies, no inference method can achieve almost everywhere convergence.
By Hanti Lin
The paper investigates fundamental limits of algorithmic principles in multiclass learning, specifically proper learning and regularization. It shows that learning cannot always be reduced to proper learning even with an enlarged hypothesis class, that proper learners may need a sublinear number of errors that can be arbitrarily large, and that regularization (SRM or local) is not universally sufficient. The authors also provide a positive theory giving sufficient conditions for SRM learnability and a characterization via integrability of revealed preferences.
By Julian Asilis, Shaddin Dughmi, Vatsal Sharan, Alec Sun, Shang-Hua Teng, Chang Wang
arXiv:2608.30246v1 Announce Type: cross
Abstract: The fundamental theorem of statistical learning states that, under suitable measurability assumptions, finite Vapnik--Chervonenkis (VC) dimension gua...
By Mateus Jesus de Arruda Campos, Gabriel Fernandes, Vinicius de Oliveira Rodrigues
arXiv:2407. 12288v5 Announce Type: replace-cross Abstract: The progress of machine learning over the past decade is undeniable.
By Hong Jun Jeon, Benjamin Van Roy
The paper presents an order-theoretic characterization of consistent inductive inference for arbitrary binary hypothesis classes. It shows that consistency—making only finitely many prediction errors on any infinite sequence labeled by an unknown hypothesis—can be captured by a single linear order on finite realizable traces. This order must satisfy two conditions: conflicting traces select different least subtraces, and the order is well‑founded on traces of each fixed target, enabling a learner whose evidence decreases with each mistake. Conversely, any consistent learner induces such an order via canonical mistake transcripts and the Kleene–Brouwer ordering.
By Zhou Lu