Saved in:
Bibliographic Details
Main Author: Khasia, Vladimer
Format: Preprint
Published: 2025
Subjects:
Online Access:https://arxiv.org/abs/2512.23448
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1866912793970606080
author Khasia, Vladimer
author_facet Khasia, Vladimer
contents Mixture of Experts (MoE) models scale capacity but often suffer from representation collapse and gradient instability. We propose Dynamic Subspace Composition (DSC), a framework that approximates context-dependent weights via a state-dependent, sparse expansion of a shared basis bank. Formally, DSC models the weight update as a residual trajectory within a Star- Shaped Domain, employing a Magnitude-Gated Simplex Interpolation to ensure continuity at the identity. Unlike standard Mixture-of-LoRAs, which incurs O(M rd) parameter complexity by retrieving independent rank-r matrices, DSC constructs a compositional rank-K approximation from decoupled unit-norm basis vectors. This reduces parameter complexity to O(M d) and memory traffic to O(Kd), while Frame-Theoretic regularization and spectral constraints provide rigorous worst-case bounds on the dynamic update. The code is available at https://github. com/VladimerKhasia/DSC
format Preprint
id arxiv_https___arxiv_org_abs_2512_23448
institution arXiv
publishDate 2025
record_format arxiv
spellingShingle Dynamic Subspace Composition: Efficient Adaptation via Contractive Basis Expansion
Khasia, Vladimer
Machine Learning
Mixture of Experts (MoE) models scale capacity but often suffer from representation collapse and gradient instability. We propose Dynamic Subspace Composition (DSC), a framework that approximates context-dependent weights via a state-dependent, sparse expansion of a shared basis bank. Formally, DSC models the weight update as a residual trajectory within a Star- Shaped Domain, employing a Magnitude-Gated Simplex Interpolation to ensure continuity at the identity. Unlike standard Mixture-of-LoRAs, which incurs O(M rd) parameter complexity by retrieving independent rank-r matrices, DSC constructs a compositional rank-K approximation from decoupled unit-norm basis vectors. This reduces parameter complexity to O(M d) and memory traffic to O(Kd), while Frame-Theoretic regularization and spectral constraints provide rigorous worst-case bounds on the dynamic update. The code is available at https://github. com/VladimerKhasia/DSC
title Dynamic Subspace Composition: Efficient Adaptation via Contractive Basis Expansion
topic Machine Learning
url https://arxiv.org/abs/2512.23448