SNR Estimation under High-dimensional Linear Models

Seminar | April 19 | 4-5 p.m. | 1011 Evans Hall

Speaker/Performer:  Xiaodong Li, UC Davis

Sponsor:  Department of Statistics

Estimation of signal-to-noise ratios and residual variances in high-dimensional linear models has important applications including heritability estimation in bioinformatics. Random effects likelihood estimators have been widely used in practice for SNR estimation, and it is known to be consistent when the model is misspecified. In this talk, we aim to investigate the conditions on both the design matrix and the coefficient vector, such that asymptotic behaviors for this SNR estimator can be explicitly derived. We will stress tools from random matrix theory and normal approximation of quadratic forms. For future work, extensions to method-of-moments, diverging aspect ratios, and linear models with feature groups will be briefly discussed. This is a joint work with my student Xiaohan Hu.

Event Contact:  CA, zhivotovskiy@berkeley.edu, 5102292370

Access Coordinator:  Nikita Zhivotovskiy,  zhivotovskiy@berkeley.edu,  510-229-2370