English

Sparse Signal Detection in Heteroscedastic Gaussian Sequence Models: Sharp Minimax Rates

Statistics Theory 2023-08-03 v4 Machine Learning Statistics Theory

Abstract

Given a heterogeneous Gaussian sequence model with unknown mean θRd\theta \in \mathbb R^d and known covariance matrix Σ=diag(σ12,,σd2)\Sigma = \operatorname{diag}(\sigma_1^2,\dots, \sigma_d^2), we study the signal detection problem against sparse alternatives, for known sparsity ss. Namely, we characterize how large ϵ>0\epsilon^*>0 should be, in order to distinguish with high probability the null hypothesis θ=0\theta=0 from the alternative composed of ss-sparse vectors in Rd\mathbb R^d, separated from 00 in LtL^t norm (t[1,]t \in [1,\infty]) by at least ϵ\epsilon^*. We find minimax upper and lower bounds over the minimax separation radius ϵ\epsilon^* and prove that they are always matching. We also derive the corresponding minimax tests achieving these bounds. Our results reveal new phase transitions regarding the behavior of ϵ\epsilon^* with respect to the level of sparsity, to the LtL^t metric, and to the heteroscedasticity profile of Σ\Sigma. In the case of the Euclidean (i.e. L2L^2) separation, we bridge the remaining gaps in the literature.

Keywords

Cite

@article{arxiv.2211.08580,
  title  = {Sparse Signal Detection in Heteroscedastic Gaussian Sequence Models: Sharp Minimax Rates},
  author = {Julien Chhor and Rajarshi Mukherjee and Subhabrata Sen},
  journal= {arXiv preprint arXiv:2211.08580},
  year   = {2023}
}
R2 v1 2026-06-28T05:59:57.766Z