High-Rank Grouped Delta Operators Enable Dynamic Geometric Transformations in Residual Networks
HRG-Delta replaces standard residual rank-1 reflections with rank-$r$ grouped subspace transformations, using dynamically generated orthonormal bases and bounded gating coefficients.
Authors
- Yonis Gulzar (ORCID: https://orcid.org/0000-0002-6515-1569)
- Mohannad Alkanan (ORCID: https://orcid.org/0000-0003-3736-7749)
- Musrrat Ali (ORCID: https://orcid.org/0000-0002-4837-1850)
- Hafed Zarzour (ORCID: https://orcid.org/0000-0001-9441-4842)
- Ammar Ladjailia (ORCID: https://orcid.org/0000-0001-8364-1791)
- Aida Bouzid (ORCID: https://orcid.org/0009-0008-0716-4709)
Institutions
- Mohamed-Cherif Messaadia University (DZ)
- King Faisal University (SA)
Publication Details
- Journal
- Zenodo (CERN European Organization for Nuclear Research)
- Published
- 2026-10-06
- DOI
- https://doi.org/10.5281/zenodo.23110502
- Primary Topic
- Advanced Neural Network Applications
- Type
- article
- Field-Weighted Citation Impact
- 0.00