Program

Preliminary, expect changes!

• Monday: Bandits & allocation · Game-theoretic statistics · Conformal prediction & forecasting · Confidence sequences & concentration · Sequential testing and adaptive inference
• Tuesday: Multiple testing & FDR · E-process optimality · Closed testing and e-closure · Conditional/sequential inference · Clinical and applied anytime-valid inference
• Wednesday: Foundations of e-values · Universal inference & likelihood ratios · Sequential testing and changepoints · Post-hoc inference & evidence synthesis · Formal methods and e-statistics · Social activity & dinner
• Thursday: Imprecise probabilities · Clinical trials and adaptive designs · Evidence, post-hoc inference & optimal e-values · Decision-making, voting and auditing · Prediction, forecasting and ML applications · Concentration, self-normalization & conditional inference
• Friday: Foundations of anytime-valid inference and future directions

Abstracts

Talks A1

Etienne Gauthier

abstract tba

See also Posters A.

Predictive and confidence regions in causal inference (Vladimir Vovk)

Joint work with Ruodu Wang.

We extend conformal e-prediction to design an algorithm for producing prediction sets in causal inference. Our algorithm works in the cases covered by the back-door criterion and those covered by the front-door criterion. However, its weakness is that it assumes that the training set is IID, or at least not too far from being IID, in some sense. In particular, our algorithm is not directly applicable in the setting of sequential decision making. To deal with this setting, we apply SAVI methods based on the law of the iterated algorithm to find explicit confidence intervals and confidence sequences. Once we have confidence intervals, we can use them to derive prediction sets; even if these prediction sets are less efficient for an IID training set, they are applicable in highly non-IID settings such as that of sequential decision making.

See also Posters A.

Brian Lee

abstract tba

See also Posters A.

Jan-Lukas Wermuth

abstract tba

See also Posters A.

Talks A2

Georgii Potapov

abstract tba

See also Posters A.

Optimal prediction with E-values (Nick Koning)

Prediction sets offer a binary inclusion / exclusion for each element at the same fixed confidence level. We generalize this to fuzzy prediction sets, which exclude elements at their own data-driven confidence level. Our key insight is that a fuzzy prediction set is equivalent to a single E-value, capturing precisely what E-values bring to prediction. Fuzzy prediction sets inherit the merging properties of their E-value, and offer richer guarantees to decision makers. We show in what sense optimal E-values give rise to optimal (fuzzy) prediction sets. We apply our results to conformal prediction, deriving optimal conformal prediction sets, and characterizing in what sense classical conformal prediction is optimal.

See also Posters A.

Talks A3

Liviu Aolaritei

abstract tba

See also Posters B.

François Caron

abstract tba

See also Posters B.

Sohom Mukherjee

abstract tba

See also Posters B.

Posters A

Wouter M. Koolen (based in part on joint work with Lukas Zierahn, Christina Katsimerou, Shubhada Agrawal and Dirk van der Hoeven)

abstract tba

Aymeric Capitaine, Antoine Scheid, Etienne Boursier, Alain Durmus, Michael I. Jordan

abstract tba

Fabian Damken, Wouter M. Koolen, Rianne de Heide

abstract tba

Guneet S. Dhillon, Javier González, Teodora Pandeva, Alicia Curth

abstract tba

Ricardo J. Sandoval, Ian Waudby-Smith, Michael I. Jordan

abstract tba

Shubhada Agrawal, Aaditya Ramdas

abstract tba

Justin Whitehouse, Zhiwei Steven Wu, Aaditya Ramdas

abstract tba

Alhad Sethi, Kavali Sofia Sagar, Shubhada Agrawal, P. N. Karthik, Debabrota Basu

abstract tba

Etienne Gauthier, Francis Bach, Michael I. Jordan

See Talks A1.

Predictive and confidence regions in causal inference (Vladimir Vovk, Ruodu Wang)

See Talks A1.

Jessica Dai, Nika Haghtalab, Brian W. Lee

See Talks A1.

Timo Dimitriadis, Jan-Lukas Wermuth and Johanna Ziegel

See Talks A1.

Georgii Potapov, Yuri Kalnishkan

See Talks A2.

Nick Koning, Sam van Meer

See Talks A2.

Talks B1

Johahnes Ruf

abstract tba

See also Posters B.

Noah Liniger

abstract tba

See also Posters B.

Qiuqi Wang

abstract tba

See also Posters B.

Jelle Goeman

abstract tba

See also Posters B.

Talks B2

Zhimei Ren

abstract tba

See also Posters B.

Yury Tavyrikov

abstract tba

See also Posters B.

Talks B3

Dante de Roos

abstract tba

See also Posters C.

Sam van Meer

abstract tba

See also Posters C.

Vikas Deep

abstract tba

See also Posters C.

Posters B

Rovanos Tsafack Nzanguim, Aurele Mingam, Jelle Goeman, Rianne de Heide

abstract tba

Aurele Mingam, Rovanos Tsafack Nzanguim, Rianne de Heide, Jelle Goeman

abstract tba

Beepul Bharti, Ambar Pal, Jeremias Sulam

abstract tba

Eugenio Clerico, Sebastian Arnold

abstract tba

Sebastian Arias, Peter Grünwald

abstract tba

D. Hop, N. Koning, S. van der Meer

abstract tba

Debolina Paul

abstract tba

Yonqqi Wang, Sebastian Arnold, Francesca Giuffrida, Peter Grünwald, Thorsten Dickhaus

abstract tba

Alhad Sethi, Kavali Sofia Sagar, Shubhada Agrawal, Debabrota Basu, P. N. Karthik

abstract tba

Liviu Aolaritei, Michael I. Jordan

See Talks A3.

Valentin Kilian, Stefano Cortinovis, François Caron

See Talks A3.

Sohom Mukherjee, Ivane Antonov, Richard Pibernik, Yo Joong Choe

See Talks A3.

Larsson, Ramdas, Ruf

See Talks B1.

Qiuqi Wang, Zhenyuan Zhang, Ruodu Wang

See Talks B1.

Noah Liniger, Ian Waudby-Smith, Antoine Scheid, Alain Durmus, Michael I. Jordan

See Talks B1.

Jelle Goeman

See Talks B2.

Yury Tavyrikov, Jelle Goeman, Rianne de Heide

See Talks B1.

Zhimei Ren

See Talks B2.

Talks C1

Lorenz Matz

abstract tba

See also Posters C.

Time-sensitive anytime-valid testing (Eugenio Clerico; joint work with Tobias Wegel, Iskander Azangulov, and Patrick Rebeschini)

Abstract: Anytime-valid tests allow evidence to be checked during data collection: one can either continue testing or stop and reject the null while still controlling type-I error. Yet, in many applications rejection is useful only if it comes soon enough. We introduce a time-sensitive testing-by-betting framework that favours early rejection by assigning rewards to rejection times and maximising their expected value under a given alternative. This encompasses hard deadlines and softer time preferences. The resulting optimal control problem admits a Bellman representation in terms only of time and evidence against the null, rather than the full history. For hard deadlines, the simple-vs-simple case reduces to a finite-horizon Neyman-Pearson problem, with a corresponding optimal e-process. Furthermore, we show that exponentially decaying rewards admit a stationary approximation, yielding the exponential-decay-optimal (EDO) criterion: a finite-time-scale counterpart to the classical growth-rate-optimal (GRO) viewpoint in anytime-valid statistics, with the GRO criterion recovered in the large-time-scale limit.

See also Posters D.

Posters C

S. Bongers, P. Grünwald

abstract tba

Rémy Degenne, Gaëtan Serré

abstract tba

Hongjian Wang

abstract tba

Dante de Roos, Peter Grünwald, Sebastian Arnold

abstract tba

Udo Boehm, Wouter Koolen, and Peter Grunwald

abstract tba

Angel Reyero Lobo, Sebastian Arias, Michele Meziu

abstract tba

Ashwin Ram, Aaditya Ramdas

abstract tba

Sam van Meer, Nick W. Koning

See Talks B3.

Vikas Deep, Shubhada Agarwal

See Talks B3.

Dante de Roos, Peter Grünwald

See Talks B3.

Lorenz Matz, Hannes Leeb

See Talks C1.

Eugenio Clerico

See Talks C1.

Talks D1

Peter Grünwald

abstract tba

See also Posters D.

Ben Chugg

abstract tba

See also Posters D.

Timothée Mathieu

abstract tba

See also Posters D.

Dennis Oestmann

abstract tba

See also Posters D.

Talks D2

Michael Lindon

abstract tba

See also Posters D.

Stef Baas

abstract tba

See also Posters D.

Talks D3

Yo Joong Choe

abstract tba

See also Posters D.

Verifying Elections with Adaptively Weighted Test Supermartingales (Alexander Ek)

Joint work with Michelle Blom, Philip B. Stark, Peter J. Stuckey, Damjan Vukcevic.

An increasingly important part of trustworthy elections is the risk-limiting audit (RLA): a statistical audit that efficiently confirms the reported outcome when it is correct or with high probability corrects the reported outcome if it is incorrect. An election audit is risk-limiting if the chance it fails to correct a wrong electoral outcome is at most a pre-specified limit.

We present recent advances in RLAs for instant-runoff voting (IRV), a ranked-choice system used in Australia, USA, and other countries. To test that the reported outcome is wrong, we need a composite hypothesis written as a union of \(O(k!)\) intersections of \(O(2^k)\) simple hypotheses when there are \(k\) candidates. Prior work addressed this by pre-selecting a subset of the simple hypotheses covering all unions. While this eliminates multiplicity, it necessitates reliable estimates of the votes cast. Instead, we use predictably weighted averages of adaptive test supermartingales, removing the need for pre-selection entirely. We call this the Adaptively Weighted Audits of Instant-Runoff Elections (AWAIRE) framework.

This work involves new statistical and computational methods in sequential anytime-valid inference. Naively testing all simple hypotheses is intractable when \(k > 6\). We take advantage of the tree structure of the tally algorithm to start by testing a few simple hypotheses, adaptively adding more as needed to confirm the results, using branch-and-bound to prune the search space. This makes it feasible to audit contests with more than 50 candidates. The approach may be useful in other situations where the null hypothesis can be expressed as a union of intersections of simple hypotheses.

See also Posters D.

Diego Martinez-Taboada

abstract tba

See also Posters D.

Posters D

Mitigating the Winner’s Curse in Dose-Ranging Studies: A Closed-Form Formula for Expected Bias and Decision Tools for Optimized Dose Selection (Victor K. de la Peña, Victor H. de La Peña, Demissie Alemayehu and Fangyuan Lin)

This paper addresses the “winner’s curse” in dose-ranging trials. In a typical Phase II study, \(K\) dose candidates are compared and the apparent winner is advanced to Phase III. Typically, the observed efficacy of the selected dose systematically overestimates its true effect. We argue that this bias is an under-appreciated contributor to the approximately 50% Phase III failure rate and to the well-documented shrinkage of treatment effects between phases. The central contribution is a closed-form expression for the expected selection bias, expressed in terms of four quantities that are readily estimable at the design stage of the pivotal trials: the number of in doses \(K\), the number of patients per arm \(n\), the outcome standard deviation \(\sigma\), and the inter-arm correlation \(\rho\):

\[\mathbb{E}[\text{Bias}] = (\sigma/\sqrt{n}) g(K) \sqrt{1 - \rho},\]

where \(g(K)\) denotes the expected value of the maximum of \(K\) standard normal random variables. We provide exact tabulated values of \(g(K)\) and show that the commonly used asymptotic approximation \(\sqrt{2 \log K}\) overstates the bias by 40%–110% for the practically relevant range \(K = 3\) to \(10\).

Extensive simulations confirm that the formula is accurate to within 0.5% across clinically realistic parameter values. To illustrate its relevance, we apply the results to several past clinical trials. We also supply several immediately applicable decision tools: a Phase III sample-size inflation factor, a calibrated go/no-go probability calculator, a cumulative-bias tracker for multi-arm multi-stage (MAMS) designs, and a dose-ranging optimizer aligned with the objectives of Project Optimus.

Joren Brunekreef, Renee Menezes, Rianne de Heide

abstract tba

Judith ter Schure

abstract tba

Rianne de Heide

abstract tba

Anurag Singh, Rajeev Verma, Julian Rodemann, Siu Lun Chau, Krikamol Muandet

abstract tba

Valentin Kilian*, Shirley Xiaoqi Liu*, Judith Rousseau, Francesco Orabona, Patrick Rebeschini

abstract tba

Yongxi Long, Alexander Ly, Nikos Ignatiadis, Jelle Goeman, Peter Grünwald, Erik van Zwet

abstract tba

Stan Koobs, Nick Koning

abstract tba

Alexander Ly, Sebastian Arias, Sebastian Arnold, Udo Boehm, Stephan Bongers, Michele Meziu, Angel Reyero Lobo, Dante de Roos, Meike Steinhilber, Yongqi Wang, and Peter Grünwald

abstract tba

Rajeev Verma, Rabanus Derr, Christian A.Naesseth, Eric Nalisnick

abstract tba

Aytijhya Saha, Aaditya Ramda

abstract tba

Sebastian Arnold*, Yo Joong (YJ) Choe*, Marco Scarsini, Ilia Tsetlin (*Equal contribution.)

See Talks D3.

Verifying Elections with Adaptively Weighted Test Supermartingales (Alexander Ek, Michelle Blom, Philip B. Stark, Peter J. Stuckey, Damjan Vukcevic)

See Talks D3.

Diego Martinez-Taboada, Aaditya Ramdas

See Talks D3.

Michael Lindon, Nathan Kallus

See Talks D2.

Stef Baas, Joost van Rosmalen, Judith ter Schure

See Talks D2.

Peter Grünwald, partially based on joint work with, Ben Chugg, Aaditya Ramdas

See Talks D1.

Ben Chugg

See Talks D1.

Timothée Mathieu, Adrienne Tuynman

See Talks D1.

Dennis Oestmann, Thorsten Dickhaus

See Talks D1.

List of Participants

• Shubhada Agrawal
• Liviu Aolaritei
• Sebastian Arias
• Sebastian Arnold
• Morgane Austern
• Stef Baas
• Beepul Bharti
• Stephan Bongers
• Bastiaan Braams
• Joren Brunekreef
• Francois Caron
• Yo Joong Choe
• Ben Chugg
• Eugenio Clerico
• Gert de Cooman
• Fabian Damken
• Vikas Deep
• Rabanus Derr
• Guneet Dhillon
• Alexander Ek
• Etienne Gauthier
• Jelle Goeman
• Peter Grünwald
• Rianne de Heide
• Roel Hulsman
• Nick Koning
• Stan Koobs
• Wouter Koolen
• Brian Lee
• Michael Lindon
• Noah Liniger
• Xiaoqi Shirley Liu
• Yongxi Long
• Nynke Luijten
• Alexander Ly
• Diego Martinez Taboada
• Timothée Mathieu
• Lorenz Matz
• Michele Meziu
• Aurèle Mingam
• Sohom Mukherjee
• Gergely Neu
• Dennis Oestmann
• Snigdha Panigrahi
• Debolina Paul
• Victor H. de la Peña
• Georgii Potapov
• Zhimei Ren
• Angel David Reyero Lobo
• Dante de Roos
• Johannes Ruf
• Ricardo Sandoval
• Antoine Scheid
• Gaëtan Serre
• Anurag Singh
• Yury Tavyrikov
• Rovanos Tsafack Nzanguim
• Adrienne Tuynman
• Dirk van der Hoeven
• Sam van Meer
• Rajeev Verma
• Vladimir Vovk
• Qiuqi Wang
• Yongqi Wang
• Jan-Lukas Wermuth
• Redouane Yagouti