Paper

On the Oracle Complexity of Interpolation-Based Gradient Descent

Unreadunread

AI summary

arXiv:2606. 19878v1 Announce Type: cross Abstract: Recent work on first-order optimizers for empirical risk minimization (ERM) has suggested that smoothness of ERM loss functions in the training data, rather than in the optimization parameters, can be leveraged to improve the oracle complexity of gradient descent (GD) methods.

Reading status

Priority

About this paper

Authors: Dongmin Lee, William Lu, Anuran Makur
Released: 2026-06-19 (1mo ago)
Topic: Related to current project
Relation to my work: Read next because On the Oracle Complexity of Interpolation-Based Gradient Descent overlaps with clean result "LoRA persona trained on <A> alone emits <B> at 23.5% when a co-trained partner learns <A>...<B>, vs 0% control on Qwen2.5-7B-Instruct (MODERATE confidence)", clean result "Leakage rate is a usable signal for recovering trigger-shaped phrases on Gaperon-1125-1B without knowing the hidden trigger itself (MODERATE confidence)", clean result "Language-mismatch LoRA SFT on Qwen2.5-7B leaks the trained completion language into bystander directives the model was never trained on, absent under same-language SFT (LOW confidence)". Matching terms: strong, line, full. Source: arxiv stat.ML (Machine Learning).
arXiv: 2606.19878

Ask Claude about this paper

Posts a comment with @claude so the runner spins up a Claude Code session scoped to On the Oracle Complexity of Interpolation-Based Gradient Descent. Replies appear in the conversation below.

Comments

No comments yet.

Related records

No links yet.