Dice Question Streamline Icon: https://streamlinehq.com

Investigate matrix e-processes

Investigate the theory of matrix e-processes, including their construction, properties, and potential advantages for composite sequential testing with matrix-valued data, beyond the initial formulations presented in this work.

Information Square Streamline Icon: https://streamlinehq.com

Background

The paper introduces matrix e-processes as a generalization of scalar e-processes, providing a framework for matrix-valued sequential testing via matrix Ville-type inequalities. While initial definitions and examples are given, a broader theory (e.g., optimality, closure properties, practical constructions, and applications) remains to be developed.

References

We also leave open some questions for future studies to address. These include a deeper understanding of the matrix randomizer U in our paper, which belongs to the "trace super-uniform" distribution family; tighter p-Chebyshev inequalities; further investigation into matrix-valued e-processes; as well as the theoretical explanation of our comparison between matrix-valued test supermartingales and scalar-valued test supermartingales when testing sequentially the same null hypotheses.

Positive Semidefinite Matrix Supermartingales (2401.15567 - Wang et al., 28 Jan 2024) in Conclusion