Watch Your Step: Adam's Margin Maximization Bias Depends on the Learning Rate Schedule
A revised arXiv paper (2611.01929v2) examines the implicit bias of the Adam optimizer and how closely it resembles sign gradient descent on homogeneous models in binary classification. The paper's headline claim is that Adam's margin maximization bias depends on the learning rate schedule. No experimental figures, model names, or datasets are given in the available abstract text, and the work is presented as an abstract-level preprint with no stated follow-up.
How this story unfolded
Key points3 items Model summary · ordered by when the evidence was published
- Point 01 11/16, 13:00
arXiv paper 2611.01929v2 is a replacement submission studying Adam's implicit bias.
Based on 1 - Point 02 11/16, 13:00
The work compares Adam with sign gradient descent on homogeneous models in binary classification.
Based on 1 - Point 03 11/16, 13:00
The paper argues Adam's margin maximization bias depends on the learning rate schedule.
Based on 1
ReportsBy publish time · newer further down
- November 16, 2026 · Mon
- Start
Watch Your Step: Adam's Margin Maximization Bias Depends on the Learning Rate Schedule
Original summary arXiv:2611.01929v2 Announce Type: replace-cross Abstract: We study the implicit bias of Adam and its similarity to sign gradient descent on homogeneous models in binary classificat
arXiv stat.ML Original ↗ - November 18, 2026 · Wed
- +2d 0h
Hybrid Quantum Selection of Learning Rates in Language-Model Training: Termination Criteria and an Interaction-Guided Schedule Oracle
Original summary arXiv:2611.21701v1 Announce Type: new Abstract: Learning-rate choices strongly affect language-model fine-tuning, but the best joint profile across layer groups is generally unknow
arXiv cs.LG Original ↗ - +2d 0h
CueSplat: Learning to Schedule Sparse Deformation for Dynamic Gaussian Splatting on Edge Devices
Original summary arXiv:2611.22095v1 Announce Type: new Abstract: Deformation-based 4D Gaussian splatting (4DGS) enables dynamic scene rendering. However, evaluating deformation networks for all Gau
arXiv cs.CV Original ↗ - Still developing · later reports will keep extending this axis