Philosophy of Science and Formal Epistemology
Research seminar series
Department of Liberal Arts
University of Technology Nuremberg
The seminars take place on Thursdays between 14.00-15.30 hours
The 50-60 min talk is followed by a 30 min discussion
and ends with tea/coffee/cookies between 15.30-16.00
Venue: Ulmestrasse 52i, Ground Floor, meeting room 02
The seminars are in hybrid format; everyone is welcome, no registration needed
Zoom link to the seminars
Organizer: Miklos Redei (m.redei@lse.ac.uk)
Programme:
- October 16, 2025
- Tom Sterkenburg (Munich Center for Mathematical Philosophy; Ludwig-Maximilians University, Munich, Germany)
- Title: "Occam's razor in machine learning"
- Abstract: The principle of Occam's razor instructs us to prefer simplicity when making inductive inferences. The principle has attracted much discussion in the philosophy of science, and it also makes a regular appearence in the machine learning literature. However, in both fields a justification for Occam's razor---why, exactly, is it a good idea to prefer simplicity?---has been elusive.
In my talk, I will first spell out a justification for Occam's razor based on the classical mathematical theory of machine learning. I argue that this is a genuine methodological justification, linking a simplicity preference to predictive accuracy, without relying on a question-begging "ontic" assumption that the "truth" is simple. I will then discuss the contemporary "generalization paradox", which appears to show that the classical theory cannot explain the performance of modern machine learning algorithms. Interestingly, attempts at a novel theoretical account to supersede the classical theory again invoke Occam's razor. I will point out, however, that this new Occam's razor does rather amount to an ontic commitment than a methodological principle.
- October 30, 2025
- Zalan Gyenis (Department of Logic, Jagiellonian University; Cracow, Poland)
- Title: "What are probabilities if logic is not classical?"
- Abstract: We would like to believe true propositions and avoid believing false ones. It is typical to represent an agent's belief state by means of a belief function which assigns numbers from the [0,1] real segment to propositions. Ideally, then, we would want our credences in true propositions to equal 1, and our credences in false propositions to equal 0. However, due to our cognitive and evidential limitations, leading to the typical human condition of imperfect information, we have to settle for something else. It is one of the basic tenets of formal epistemology that belief functions of a rational agent are weighted means of classical truth evaluations; this is the same as saying that they belong to the ``convex hull'' of classical evaluations, and, seen from yet another angle, it means that these belief functions satisfy the classical Kolmogorov probability axioms.
All this assumes, usually implicitly, that the underlying logic is classical. How does the situation change if this assumption is removed? The talk concerns with axiomatization of probability in the context of various nonclassical logics. First, I survey the known results, then I report a solution to an open problem regarding the axiomatization of probability functions of Symmetric Logic. Finally, I show that a large class of general logics admit a recursive axiomatization of their probability functions, allowing probabilistic reasoning in an axiomatic manner. I close the talk with mentioning open problems and possible research future directions.
- November 20, 2025
- Iulian Toader
- Title: "The roots of scientific conservatism"
- Abstract: TBA
- November 27, 2025 (online only seminar)
- December 4, 2025
- Gabor Hofer-Szabo (ELTE Research Centre for the Humanities, Budapest, Hungary)
- Title: "Classicality, locality and the three types of Bell's inequalities"
-
Abstract: Bell's inequalities can be understood in three different ways depending on whether the numbers featuring in the inequalities are interpreted as classical probabilities, classical conditional probabilities, or quantum probabilities. In the talk I will argue that the violation of Bell's inequalities has different meanings in the three cases. In the first case it rules out the interpretation of certain numbers as probabilities of classical events. In the second case it rules out a common causal explanation of conditional correlations of certain events (measurement outcomes) conditioned on other events (measurement settings). Finally, in the third case the violation of Bell's inequalities neither rules out the interpretation of these numbers as probabilities of events nor a common causal explanation of the correlations between these events---provided both the events and the common causes are interpreted non-classically.