Dynamic Subspace Composition: Efficient Adaptation via Contractive Basis Expansion
This is an incremental improvement for efficient adaptation in large-scale machine learning models.
The paper tackles the issues of representation collapse and gradient instability in Mixture of Experts models by proposing Dynamic Subspace Composition, which reduces parameter complexity from O(M rd) to O(M d) and memory traffic to O(Kd).
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