English

Tessellated Distributed Computing

Information Theory 2024-04-23 v1 math.IT

Abstract

The work considers the NN-server distributed computing scenario with KK users requesting functions that are linearly-decomposable over an arbitrary basis of LL real (potentially non-linear) subfunctions. In our problem, the aim is for each user to receive their function outputs, allowing for reduced reconstruction error (distortion) ϵ\epsilon, reduced computing cost (γ\gamma; the fraction of subfunctions each server must compute), and reduced communication cost (δ\delta; the fraction of users each server must connect to). For any given set of KK requested functions -- which is here represented by a coefficient matrix FRK×L\mathbf {F} \in \mathbb{R}^{K \times L} -- our problem is made equivalent to the open problem of sparse matrix factorization that seeks -- for a given parameter TT, representing the number of shots for each server -- to minimize the reconstruction distortion 1KLFDEF2\frac{1}{KL}\|\mathbf {F} - \mathbf{D}\mathbf{E}\|^2_{F} overall δ\delta-sparse and γ\gamma-sparse matrices DRK×NT\mathbf{D}\in \mathbb{R}^{K \times NT} and ERNT×L\mathbf{E} \in \mathbb{R}^{NT \times L}. With these matrices respectively defining which servers compute each subfunction, and which users connect to each server, we here design our D,E\mathbf{D},\mathbf{E} by designing tessellated-based and SVD-based fixed support matrix factorization methods that first split F\mathbf{F} into properly sized and carefully positioned submatrices, which we then approximate and then decompose into properly designed submatrices of D\mathbf{D} and E\mathbf{E}.

Keywords

Cite

@article{arxiv.2404.14203,
  title  = {Tessellated Distributed Computing},
  author = {Ali Khalesi and Petros Elia},
  journal= {arXiv preprint arXiv:2404.14203},
  year   = {2024}
}

Comments

65 Pages, 16 figure. The manuscript is submitted to IEEE Transactions on Information Theory