About
Welcome to the inaugural conference for the ERC-funded Epistemic Utility for Imprecise Probability project.
The project aims to characterise reasonable measures of accuracy for imprecise probabilities and to use these measures to justify and extend imprecise Bayesian methods.
This inaugural event covers a wide range of topics in imprecise probability theory. Its primary aim is to bring together two communities: the IP community and the philosophical community.
The conference was held virtually via Zoom. You can find the recordings in the schedule below.
We would like to thank the audience, and our eight invited speakers for their thought-provoking presentations!
News: We organise the SIPTA Summer School 2022, taking place 15 – 19 August in Bristol.
Speakers
-
Catrin Campbell-Moore Thursday, 21 October, 12:15 – 13:30
Lecturer in Philosophy at the University of Bristol
-
Jasper De Bock Thursday, 21 October, 13:45 – 15:00
Professor at the Foundations Lab for imprecise probabilities of Ghent University
-
Snow Zhang Thursday, 21 October, 16:15 – 17:30
Bersoff Faculty Fellow at New York University
-
Teddy Seidenfeld Thursday, 21 October, 17:45 – 19:00
Herbert A. Simon University Professor of Philosophy and Statistics at Carnegie Mellon University
-
Seamus Bradley Friday, 22 October, 12:15 – 13:30
Marie Curie Individual Fellow at the University of Leeds
-
Gert de Cooman Friday, 22 October, 13:45 – 15:00
Professor in Uncertainty Modelling and Systems Science at the Foundations Lab for imprecise probabilities of Ghent University
-
Ruobin Gong Friday, 22 October, 16:15 – 17:30
Assistant Professor of Statistics at Rutgers University.
-
Miriam Schoenfield Friday, 22 October, 17:45 – 19:00
Associate Professor in the Department of Philosophy at The University of Texas at Austin
Schedule
The presentations will be 45 minutes, followed by a discussion of 30 minutes. There will be a break between 15:15 and 16:15.
All times are in British Summer Time.
| Time | Slot | Title |
|---|---|---|
| Thursday, 21 October, 12:15 – 13:30 |
Catrin Campbell-Moore
University of Bristol
|
Beliefs as Probability Constraints slides recording Click to view abstractI propose capturing our epistemic states by a collection of probability constraints. For example, it might contain the constraint that pr(Rain)>0.7. This naturally allows for imprecision when you are not fully opinionated: you leave open whether pr(Rain)>0.7 or not. The model is expressively powerful, allowing us ot capture a range of epistemic states considered in the imprecise probability literature such as a set of desirable gambles and (some aspects of) choice functions. |
| Thursday, 21 October, 13:45 – 15:00 |
Jasper De Bock
Ghent University
|
The meaning of imprecise probabilities: an axiomatic perspective based on choice functions and desirability slides recording |
| Thursday, 21 October, 16:15 – 17:30 |
Snow Zhang
New York University
|
Pooling and updating on experts' opinions with imprecise probabilities slides recording Click to view abstractWe frequently consult experts for their opinions on their subjects of expertise. In this context, two questions arise: (i) how different experts' opinions should be aggregated into a single quantity representative of their collective opinion as a group, and (ii) how a rational non-expert should revise their own opinion, given multiple experts' opinions. In both cases, certain plausible principles turn out to be jointly incompatible with each other within the Bayesian paradigm. However, recently many (e.g. Stewart and Ojea Quintana 2017, 2018; Elkin and Wheeler, 2018) suggest that we get around the impossibility results with imprecise probabilities. In this talk I give one set of conditions under which imprecise probability is the only way to avoid the impossibility results. I discuss how this result sheds new light on the problems of pooling and updating on experts' opinions. |
| Thursday, 21 October, 17:45 – 19:00 |
Teddy Seidenfeld
Carnegie Mellon University
|
Some strategic aspects of forecasting with strictly proper scoring rules slides recording Click to view abstractDe Finetti's (2-person, 0-sum, sequential) Prevision Game for buying and selling random variables – the so-called "Book" game for betting on events when restricted to buying and selling indicator variables – provides the context for his seminal result about coherent pricing. The Merchant plays first and is obliged to fix 2-sided, fair prices at which the Customer, who plays second, may buy or sell random variables with the Merchant. The Merchant's prices for random variables are defined as coherent iff the Customer cannot create (finitely many) contracts at the Merchant's prices that produce a (uniform) sure gain; hence, a uniform sure loss for the Merchant. De Finetti established that a set of prices for random variables are coherent if and only if there exists a (finitely additive) expectation function for random variables that agrees with those prices. (This reasoning extends to matching coherent called-off prices, given an event B, with conditional expectations given B.) However, even in the case of pricing bets on events, de Finetti's result does not establish that the Merchant will announce coherent prices that agree with her/his personal probabilities for the corresponding events: It is well known there are opportunities for the Merchant to play strategically, rather than straightforwardly, when announcing prices. Moreover, the strategic play may make it impossible to reverse engineer the Merchant's straightforward expectations. Then, coherent pricing does not succeed in operationalizing de Finetti's theory of Personal Probability. In order to robustify his operational theory of coherence against such concerns of strategic play, de Finetti switched from the 2-person Prevision Game to a 1-person Forecasting Game: using Brier-score to assess forecasts. Brier-score, a strictly proper scoring rule, is offered as incentivizing straightforward play by the Forecaster. In this presentation I review some residual aspects of strategic play in de Finetti's Forecasting Game. These opportunities for strategic forecasting result from the need to choose a unit (a currency) for realizing the Brier-score. (A detailed account of this problem is discussed in Section 6 of our (2013) paper, Exchange Rates, Phil. Sci. 80, No. 4: 504-532.) I conclude the presentation by noting that these strategic aspects of forecasting with a strictly proper scoring rule apply, also, to the interpretation of scoring rules as epistemic utilities. That is, these aspects of strategic forecasting pose a challenge to formalizing the epistemic goal of accuracy using scoring rules applied to forecasting. |
| Friday, 22 October, 12:15 – 13:30 |
Seamus Bradley
University of Leeds
|
How to be an Imprecise Impermissivist handouts recording |
| Friday, 22 October, 13:45 – 15:00 |
Gert de Cooman
Ghent University
|
Randomness, imprecision and scoring Flipping a few ideas around slides recording |
| Friday, 22 October, 16:15 – 17:30 |
Ruobin Gong
Rutgers University
|
Understanding the Epistemic Utility of Imprecise Probabilities in Statistical Inference slides recording Click to view abstractImprecise probabilities have many good uses in statistical inference. The analyst may not know what prior to use for a Bayesian model, what mechanism gave rise to the missing data, or how to make probabilistic statements when non-identifiable parameters are involved. Such kinds of uncertainty are structurally intrinsic to the statistical model, and imprecise probabilities can well articulate them without concocting unwarranted assumptions. On the other hand, imprecise probabilities present unique challenges that call for the judicious judgment on the analyst's part. The plurality of updating rules leads to seemingly paradoxical phenomena such as dilation and sure loss. In addition, while their results are more robust and informative, IP models are generally difficult to compute. In this talk, I deliberate the benefits and difficulties with imprecise probabilities in statistical inference, using a few examples encountered in practice. I call for a principled method to understand the epistemic utility in the statistical application of imprecise probabilities. |
| Friday, 22 October, 17:45 – 19:00 |
Miriam Schoenfield
University of Texas at Austin
|
Bringing the Permissivism Debate Back Down to Earth recording |
Attending the conference
All the presentations will be broadcast via Zoom. Everyone is welcome to follow the presentations and be involved in the discussions.
The conference will cover a broad range of topics of interest to both philosophers and researchers in the imprecise probabilities community. To register for free, please fill out this form.
Registration is required. We will use your information to distribute the Zoom link and password for the presentations. We are looking forward to seeing you virtually at our conference!
Contact
For more information please contact Jason Konek, Kevin Blackwell or Arthur Van Camp.