日本語

Workshop papers

The Geometry of Spectral Gradient Descent: Layerwise Criteria for SignSGD vs SpecSGD

Hiroki Naganuma (equal contribution), Laura Gomezjurado Gonzalez (equal contribution), Mahdi Ghaznavi (equal contribution), Ioannis Mitliagkas

Equal contribution

ICLR 2026 Workshop on Geometry-grounded Representation Learning and Generative Modeling · February 2026

OpenReview
lDs4vkSX7J

Research record

This page provides the canonical citation and related research links. See the original record for the abstract.

Highlights

  • Oral Presentation
  • Travel Grant

Research links

How to cite

Hiroki Naganuma, Laura Gomezjurado Gonzalez, Mahdi Ghaznavi, Ioannis Mitliagkas. “The Geometry of Spectral Gradient Descent: Layerwise Criteria for SignSGD vs SpecSGD.” ICLR 2026 Workshop on Geometry-grounded Representation Learning and Generative Modeling, 2026.

@inproceedings{Naganuma2026The,
  title = {The Geometry of Spectral Gradient Descent: Layerwise Criteria for SignSGD vs SpecSGD},
  author = {Hiroki Naganuma and Laura Gomezjurado Gonzalez and Mahdi Ghaznavi and Ioannis Mitliagkas},
  year = {2026},
  booktitle = {ICLR 2026 Workshop on Geometry-grounded Representation Learning and Generative Modeling},
  url = {https://openreview.net/forum?id=lDs4vkSX7J}
}