Abstract
Given a heterogeneous Gaussian sequence model with unknown mean and known covariance matrix , we study the signal detection problem against sparse alternatives, for known sparsity s. Namely, we characterize how large should be, in order to distinguish with high probability the null hypothesis from the alternative composed of s-sparse vectors in , separated from 0 in norm () by at least . We find non-asymptotic minimax upper and lower bounds over the minimax separation radius 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 with respect to the level of sparsity, to the metric, and to the heteroscedasticity profile of Σ. In the case of the Euclidean (i.e. ) separation, we bridge the remaining gaps in the literature.
Acknowledgements
We would like to thank the anonymous Associate Editor and Referee for many suggestions that helped improve the present manuscript, and the anonymous Referee for pointing out the explicit expression of from Theorem 4.
Citation
Julien Chhor. Rajarshi Mukherjee. Subhabrata Sen. "Sparse signal detection in heteroscedastic Gaussian sequence models: Sharp minimax rates." Bernoulli 30 (3) 2127 - 2153, August 2024. https://doi.org/10.3150/23-BEJ1667
Information