EPIMP Concluding Conference 2024

18 and 19 December 2024, University of Bristol

About

Welcome to the EPIMP Concluding Conference 2024!

Welcome to the concluding 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 concluding 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.

Nikos Gamez–Konek     

"Imprecise probabilities are the only treats I need."
— Jason Konek, 2021












Finley Gamez–Konek     

"Trying to measure my excitement level is like using imprecise probabilities. You might end up with "probably barking" or "almost-surely tail-wagging" when in reality, it's a full-blown chaos of joy and fur."
— Finley Gamez-Konek (free translation from Barklish), 2024






Organising committee

Jason Konek (jason.konek@bristol.ac.uk), chair
Kevin Blackwell (kevin.blackwell@bristol.ac.uk)
Giacomo Molinari (giacomo.molinari@bristol.ac.uk)
Arthur Van Camp (a.van.camp@tue.nl)

The organisation of this conference is made possible by Jason Konek's ERC-funded Epistemic Utility for Imprecise Probability project.

European Research Council Epistemic Utility for Imprecise Probability project University of Bristol

Location

Old Council Camber Old Council Chamber






Wills Memorial Building, Queens Rd, Bristol BS8 1QE, United Kingdom


View larger map

Schedule

Here is the tentative schedule. Please note that it may be subject to changes.

The presentations will be 45 minutes, followed by a discussion of 30 minutes.

Time Speaker Title
Wednesday
18 December
9:00 – 10:15		 Francesca Zaffora Blando Carnegie Mellon University How to de-idealize: Computability and Imprecision
Click to view abstract
Bayesian rationality is demanding: in the classical setting, Bayesian reasoners are required to be logically omniscient and to assign sharp credences to every measurable event, and they are assumed to be computationally unbounded. Since some of these demands are plausibly beyond the reach of actual reasoners like us, in recent years Bayesian epistemology has been going through a general project of de-idealization. In this talk, I will explore some of the consequences of de-idealizing Bayesian epistemology: in particular, some of the consequences of relaxing the classical assumptions that credences ought to always be precise and that Bayesian reasoners are unencumbered by computational limitations. I will do so by addressing traditional issues in Bayesian epistemology (such as convergence to the truth, intersubjective agreement, and polarization of opinions) through the prism of computable probability theory and by drawing on recent work that connects Bayesian epistemology with the theory of algorithmic randomness—a branch of computability theory that specifies what it takes for an outcome of a probabilistic process to be typical in an algorithmically specifiable way. Algorithmic randomness was recently generalized to the setting of imprecise probability, so my aim in this talk is for our discussion to serve as a springboard for extending current work in computable probability theory and computable Bayesian epistemology to the imprecise probability setting.
Wednesday
18 December
10:30 – 11:45		 Branden Fitelson Northeastern University Elimination Counterexamples: State of the Art
Click to view abstract
This talk will analyze the recent dialectic concerning Elimination Counterexamples (ECEs) to various strictly proper scoring rules. After rehearsing some historical background on ECEs, I will focus on two recent strategies for responding to such counterexamples (on behalf of veritistic Bayesian epistemology).
Wednesday
18 December
13:00 – 14:15		 Jason Konek University of Bristol
Giacomo Molinari University of Bristol 		An Accuracy Argument for Self Trust
Wednesday
18 December
14:30 – 15:45		 Gregory Wheeler Frankfurt School of Finance & Management Beyond Expectation
Click to view abstract
The desirable gambles framework is the foundation for lower previsions, which includes de Finetti's linear previsions. These previsions serve as a cornerstone for practical applications in decision theory and uncertainty modeling, particularly in areas such as financial risk assessment and Bayesian inference. They also connect to broader theoretical advancements, providing a bridge between classical and imprecise probabilities. The elegance of the axioms and the staggering body of results provide the best unifying theory of imprecise probability. The first part of this talk celebrates the desirable gambles framework and its success, a success that has been arguably beyond expectations.

The second part of the talk examines the roles of mathematical expectation and linear functionals in the desirable gambles framework, exploring why and how these conditions might be weakened. Through these examples, I argue that expectations, while useful, are insufficient for capturing the full richness of probabilistic reasoning, especially in dynamic, high-dimensional, or decision-theoretic contexts. This insufficiency highlights the need for alternative frameworks or extensions that better account for these complexities. This calls for embracing the pragmatism at the heart of Bayesianism (broadly construed), resisting reification, and developing more flexible tools for reasoning under uncertainty.
Wednesday
18 December
16:00 – 17:15		 Gert de Cooman Ghent University Unifying classical and quantum probability: it's all about symmetries
Click to view abstract
A number of research groups have started to pay attention to the connections between quantum theory and imprecise probabilities, and the aim of this lecture is to lay out the basic ideas that make this connection interesting and appealing. I start with a discussion of an abstract theory of desirability on general option spaces, which allows us to point out the simple basic principles that underlie conservative probabilistic inference, and which leads directly to the resulting derived notions of coherent (lower and upper) previsions. I then show how simple symmetry considerations allow us to derive both classical probabilities and quantum probabilities as special instances.
Thursday
19 December
9:00 – 10:15		 Matthias Troffaes Durham University Structured Elicitation of Epistemic Uncertainty in Decision Problems
Click to view abstract
When eliciting epistemic uncertainty from an expert, using probability bounds provides a very flexible way for the expert to express their uncertainty. Additionally, it has been argued that bounding encourages the expert to be clear and honest about any severe uncertainties they may have. Unfortunately, eliciting probability bounds is more complex than eliciting exact probabilities, as there are many more models to choose from. However, when we aim to solve a specific decision problem, it is usually not needed to completely identify the expert's bounds on all random quantities. In this presentation, I will present and visualize a new theoretical result concerning the range of coherent extensions of such bounds. I will then use this result to demonstrate how we can enable elicitation of uncertainty to solve a specific decision problem, using as few queries as possible. This reduces the cognitive burden on the expert, thereby simplifying elicitation procedures for bounded probabilities.

Joint work with Nawapon Nakharutai & Sébastien Destercke.
Thursday
19 December
10:30 – 11:45		 Catrin Campbell-Moore University of Bristol Probability filters
Click to view abstract
We discuss probability filters as a model of uncertain belief, capturing an agent’s epistemic state with a collection of probability constraints which should be closed under finite entailment. This offers a natural imprecise probability model which is also very flexible, allowing it to accommodate infinitesimals in an intuitive way. We discuss how to interpret it and what epistemic methods may be obtained.
Thursday
19 December
13:00 – 14:15		 J. Dmitri Gallow University of Southern California Structured Deliberation
Click to view abstract
Rather than deciding between all of the options on the menu in one fell swoop, you could instead structure your deliberation by splitting your decision up into smaller sub-decisions. First, decide whether to order a chicken or fish dish. Then, decide which dish of the chosen category to order. That is, rather than deliberating about the synchronic, all-at-once decision you actually face, you could instead deliberate about a hypothetical diachronic decision with two choice-points: first, choose a submenu, and next, choose an option from that submenu. In this talk, I'll defend two theses about this kind of structured deliberation. The first thesis is that structured deliberation can lead you to dismiss a perfectly rational option for bad reasons. The second thesis is that we nonetheless do structure our deliberation about complicated decisions in this way. For this reason, our intuitions about complicated decisions are susceptible to a novel kind of framing effect. I apply this lesson to some recent decisions discussed by Jack Spencer, Ian Wells, and Joe Horton.
Thursday
19 December
14:30 – 15:45		 Kevin Blackwell University of Bristol
Arthur Van Camp Eindhoven University of Technology 		talk 1 Marginal extension for choice functions
Click to view abstract
Consider two uncertain variables X and Y, each taking values in a finite possibility space. X comes with marginal information in the form of a belief model on X, and Y comes with conditional information in the form of a belief model on Y conditional on every non-empty event about X. The marginal extension of X and Y, if it exists, is a belief model on X and Y that agrees with both the marginal and conditional information. We show that, if the belief models are coherent choice functions, the marginal extension always exists, and there is a unique smallest one.

talk 2 Coherent Rejection Functions for Arbitrary Things
slides
Click to view abstract
This mini-talk will be an exercise in machinery-building. The question is: how do we characterize (axiomatize) coherent rejection functions for arbitrary objects? The first insight is that we can represent binary comparisons of objects by ordered pairs – which are just another kind of thing; it is simple to represent coherent preference orders directly in terms of these higher-order objects. Once we have coherence axioms for sets of desirable things of this kind, we can leverage the sets of desirable sets of things results of Gert de Cooman, Arthur Van Camp, and Jasper De Bock to immediately see what the corresponding coherence axioms are for SDS of these things. But these are not in one-to-one correspondence with rejection functions for the original things; they express more. The second insight is that rejection functions do correspond exactly to “set preferences”. So I give coherence axioms for set preferences, which equivalently fully characterizes coherence for rejection functions. I present two main results: (1) the theorem which shows that a set preference satisfies my coherence axioms if and only if it satisfies the intuitive notion of coherence (cashed out in terms of a coherent dominating binary preference order representation). (2) Finally, we can represent coherent rejection functions for the original things in terms of correspondence to a particular subset of a coherent SDS for the ordered-pair things.
Thursday
19 December
16:00 – 17:15		 Kevin Dorst Massachusetts Institute of Technology Ambiguity Drives Hindsight Bias
Click to view abstract
Some of our uncertain judgments are clear—How likely is a fair coin to land heads? Others are ambiguous—How likely am I to own a dozen spoons? Ambiguous judgments generate higher-order uncertainty: we are uncertain what our own subjective probabilities are. We can model higher-order uncertainty using variants of Harsanyi type spaces, generalizing Hintikka models of epistemic logic. Using them, I argue that the contrast between clarity and ambiguity helps explain hindsight bias: the tendency for learning something to lead us to increase our estimate for our prior probability for it. (“I knew it all along.”) I’ll show that for Bayesians, hindsight bias is rational when and only when (1) your prior credence is ambiguous and (2) you trust your priors. Two new experiments confirm these predictions.

Speakers

  • Francesca Zaffora Blando Wednesday, 18 December, 9:00 – 10:15

    Assistant Professor in the Department of Philosophy at Carnegie Mellon University

  • Branden Fitelson Wednesday, 18 December, 10:30 – 11:45

    Distinguished Professor of Philosophy at Northeastern University

  • Jason Konek Wednesday, 18 December, 13:00 – 14:15

    Senior Lecturer in philosophy at the University of Bristol

  • Giacomo Molinari Wednesday, 18 December, 13:00 – 14:15

    Postdoctoral researcher on the EpImp project at the University of Bristol

  • Gregory Wheeler Wednesday, 18 December, 14:30 – 15:45

    Professor of Philosophy and Computer Science and Head of the Department of Computational Science & Philosophy at the Frankfurt School of Finance & Management

  • Gert de Cooman Wednesday, 18 December, 16:00 – 17:15

    Senior Full Professor of Uncertainty Modelling and Systems Science at Ghent University and Honorary Professor at the University of Bristol’s Department of Philosophy

  • Matthias Troffaes Thursday, 19 December, 9:00 – 10:15

    Professor of Probability in the Department of Mathematical Sciences at Durham University

  • Catrin Campbell-Moore Thursday, 19 December, 10:30 – 11:45

    Senior Lecturer in Philosophy at the University of Bristol

  • J. Dmitri Gallow Thursday, 19 December, 13:00 – 14:15

    Associate Professor of Philosophy at the University of Southern California

  • Kevin Blackwell Thursday, 19 December, 14:30 – 15:45

    Lecturer in Philosophy at the University of Bristol

  • Arthur Van Camp Thursday, 19 December, 14:30 – 15:45

    Assistant Professor at Eindhoven University of Technology

  • Kevin Dorst Thursday, 19 December, 16:00 – 17:15

    Associate Professor in the Department of Linguistics and Philosophy at MIT

Attending the conference

This event is open to everyone interested in imprecise probabilities. Registration is free of charge, and we will provide lunch and coffee to all registered participants. To secure your place, please complete this Doodle form by 7 December.
Speakers do not need to complete this form.

Travel from Bristol Airport

  • Bus: Bristol Airport to Bristol Temple Meads via Airport Flyer; Bristol Temple Meads to Wills Memorial Building by local bus (number 8, 71 or 72) to Park Street Top (U6).

  • Taxi: directly from the airport to Wills Memorial Building.

Travel from Heathrow

Contact

For more information please contact Jason Konek.
For comments about the website, please contact Arthur Van Camp.