For the last 5 years, I have been interested in a variety of topics in multiple hypothesis testing, and related topics in selective inference (post-selection inference), simultaneous inference, etc.
Online multiple testing considers the setting in which a sequence of decisions has to be made one at a time in an online fashion. Usually, these decisions are tests of different hypotheses, to either test a global null hypothesis or control the familywise error rate or false discovery rate. However, one may instead be interested in constructing a series of confidence intervals, or just making sign decisions on parameters, and controlling the false coverage rate or false sign rate.
ADDIS: an adaptive discarding algorithm for online FDR control with conservative nulls J. Tian, A. RamdasNeurIPS, 2019arxivcode
Online control of the false coverage rate and false sign rate A. Weinstein*, A. Ramdas*ICML, 2020arxivproc
Online control of the familywise error rate J. Tian, A. RamdasStatistical Methods in Medical Research, 2021arxivproc
Asynchronous online testing of multiple hypotheses T. Zrnic, A. Ramdas, M. JordanJMLR, 2021arxivcodeblogproc
The power of batching in multiple hypothesis testing T. Zrnic, D. Jiang, A. Ramdas, M. JordanAISTATS, 2020arxivtalkproc
SAFFRON: an adaptive algorithm for online FDR control A. Ramdas, T. Zrnic, M. Wainwright, M. JordanICML, 2018arxivprocgithub (full oral talk)
MAB-FDR: Multi (A)rmed/(B)andit testing with online FDR control F. Yang, A. Ramdas, K. Jamieson, M. WainwrightNeurIPS, 2017arxivcode30-min talkproc (spotlight talk)
Online control of the false discovery rate with decaying memory A. Ramdas, F. Yang, M. Wainwright, M. JordanNeurIPS, 2017arxivsummaryproc15-min talk from 44:00 (full oral)
Dynamic Algorithms for Online Multiple Testing Z. Xu, A. RamdasMathematical and Scientific ML, 2021arxiv
Interactive/dynamic multiple testing
I'm interested in the question of how to involve a human-in-the-loop while testing one or multiple hypotheses. These new types of procedures are “interactive”, and involve the human making a sequence of (arbitrary, but hopefully smart) data-dependent decisions. The aim is two-fold: (a) no matter what the human does, type-1 error should be controlled, (b) if the human made smart decisions (based on the data, and any heuristic model), the power should be higher than what one can achieve with a noninteractive test.
Interactive identification of individuals with positive treatment effect while controlling false discoveries B. Duan, L. Wasserman, A. Ramdasarxivtalk
STAR: A general interactive framework for FDR control under structural constraints L. Lei, A. Ramdas, W. FithianBiometrika, 2020arxivproccodetalk
Familywise error rate control by interactive unmasking B. Duan, A. Ramdas, L. WassermanICML, 2020arxivcode
Interactive martingale tests for the global null B. Duan, A. Ramdas, S. Balakrishnan, L. WassermanElectronic J. of Stat., 2020arxivcodeproc
Interactive rank testing by betting B. Duan, L. Wasserman, A. RamdasCausal Learning and Reasoning, 2022arxiv
Simultaneous high-probability bounds on the FDP in structured, regression and online settings E. Katsevich, A. RamdasAnnals of Stat., 2020arxivproccode
Large-scale simultaneous inference under dependence J. Tian, X. Chen, E. Katsevich, J. Goeman, A. Ramdasarxiv
Structured multiple testing
There is much interest in how to incorporate various notions of structure into multiple testing to improve both power and precision of off-the-shelf procedures. I have considered various different types of “structure” in my past work, but it seems like there is much that remains to be explored.
E-values as unnormalized weights in multiple testing R. Wang, A. Ramdasarxiv
False discovery rate control with e-values R. Wang, A. RamdasJ of Royal Stat. Soc., Series B, 2022arxiv
A unified framework for bandit multiple testing Z. Xu, R. Wang, A. RamdasNeurIPS, 2021arxiv
Post-selection inference for e-value based confidence intervals Z. Xu, R. Wang, A. Ramdasarxiv
A unified treatment of multiple testing with prior knowledge using the p-filter A. Ramdas, R. F. Barber, M. Wainwright, M. JordanAnnals of Stat., 2019arxivcodeproc
DAGGER: A sequential algorithm for FDR control on DAGs A. Ramdas, J. Chen, M. Wainwright, M. JordanBiometrika, 2019arxivcodeBM
p-filter: multi-layer FDR control for grouped hypotheses R. F. Barber*, A. Ramdas*J of Royal Stat. Soc., Series B, 2016arxivcodeJRSSB
QuTE: decentralized FDR control on sensor networks A. Ramdas, J. Chen, M. Wainwright, M. JordanIEEE CDC, 2017codeCDC
Miscellaneous
Fast and powerful conditional randomization testing via distillation M. Liu, E. Katsevich, L. Janson, A. RamdasBiometrika, 2021arxivcode
Optimal rates and tradeoffs for multiple testing M. Rabinovich, A. Ramdas, M. Wainwright, M. JordanStat. Sinica, 2019arxivproc