Workshop on Model independence in Physics
This talk begins from the view that physics aims not only to provide accurate predictions and descriptions of the world, but to explain it. The turn towards model independence is not a step forward in this endeavour. In searching for new physics with model independent methods, such as precision measurements and bottom-up EFTs, one is merely looking for deviations and to parameterise new physics, but not to present new particles, symmetries, and interactions that may be used to provide genuine explanations of phenomena. The paper presents this challenge and contrasts these bottom-up EFTs with top-down EFTs, which may under certain circumstances be considered to be explanatory.
Effective field theories are convenient to systematically explore quantum field theories in the presence of a large separation of physical scales, be there the weak scale and the scale of new physics. EFTs can be matched to any specific UV physics scenario but they can also be used to parametrise unknown physics in terms of few coefficients capturing local interactions of know particles. I shall discuss the applications of EFTs in the analyses of LEP and LHC data. I shall also mention some theoretical constraints following from the principle of locality and unitarity. Consistency with the laws of gravity at the quantum level also imposes some non-trivial constraints on these IR coefficients, delineating the borders of the swampland.
Throughout frontier physics, the lack of direct, easily accessible empirical evidence has problematized traditional theory construction methods---in particular straightforward model construction. The best-motivated models that go beyond known physics are either untestable or disfavoured by the lack of discovery of new evidence. But theory construction has not stopped; instead, I argue that a new methodology of framework generalization has taken hold in many areas of frontier physics. I highlight this method in particle physics, quantum foundations, and gravitational physics to highlight the ubiquity of this method. I highlight some of its benefits, like turning precision testing into exploratory experimentation, as well as the limitations. Even if model building is the final goal, we can go beyond models to deal with the unique circumstances facing frontier physics today.
I will discuss impossibility / no-free lunch results relevant to anomaly detection / out of distribution detection. I will also discuss the role of bias and priors and examine the spectrum of model dependence in HEP searches.
One important element in assessing the adequate balance between model independent search and guidance by models is the epistemic significance of support for models. The present paper aims to develop a conceptual framework for assessing that question by developing a Bayesian model of belief updating under the most relevant observations. The model will be discussed in the context of low energy SUSY. The goal is not to come up with any specific quantification of the currently justified degree of trust in low energy SUSY. Rather, the paper aims to identify structures of reasoning and salient connections between prior assumptions and resulting degrees of trust.
Discussing the strategies of model-independent searches is closely related to the question of minimal requirements to be model-free but still able to grasp hypothetical (new) discoveries appropriately. One can start from traditional operationalism, which allowed concept formulation by using experimental operations only. My attempt is to show that an extension, which uses basic theoretical assumptions that can later be discarded, results in so-called operational entities as useful tools for model-independence.
While this workshop explores how theory and experimentation can be independent from models, my work (Pirtle et al 2018) explores a related question: how can models be independent from one another? Using case studies and Richard Levins’ framework on model pluralism, I define a series of dimensions by which models can be independent from one another, allowing for more fine-grained discussions on how to design and choose models for specific analytical tasks.
I will present my views on the recent evolution of beyond the Standard Model model building, its impact on experimental search strategies and on the development of a plan for the future of high energy physics.
Bottom-up EFTs have taken an increasingly central place in particle physics since the early 1980s, opening, in particular, new ways to learn more about BSM physics without relying on specific high-energy models. Despite this growing importance, their historical development remains largely uncharted so far. In this talk, I trace the origin of bottom-up EFTs, identifying some of their most significant historical roots in Steven Weinberg’s works in the 1960s (among others). Although his power-counting concept is rudimentary at this stage, we can still identify (or so I will argue) key defining features of bottom-up EFTs in these early works. I will conclude with a few remarks about Weinberg’s early notion of model independence in this context.
Much of the literature on scientific models is concerned with representational models. Toy models, on the other hand, (a kind of nonrepresentational model) are rarely discussed. This paper aims to elevate the status of toy models: by (1) discussing the connection between user intentionality, toy models and scientific representation, by (2) highlighting a way in which the Ehrenfests’ urn model has been used to reason about physical theories, and by (3) explaining why it can be used in this way without performing a representational function.
This poster purports to examine the epistemology of Model-independent approaches to dark matter detection by looking at various methods in the literature, both experimental and theoretical. The case of DM detection presents an interesting case study as any prior knowledge of DM comes from galactic models. These methods will be framed philosophically through the lens of Quinean naturalised epistemology and different modelling approaches.
When do machine learning models provide understanding of phenomena? In this talk, I argue that ML models can aid understanding of phenomena when there is low 'link uncertainty.' I discuss how this may differ from gaining understanding from non-ML models.
There exists a strong contrast between experimental and theoretical motivations for physics beyond the Standard Model of particle physics on the one hand, and concrete observations of new particles at the LHC on the other one. Combined with recent progress in machine learning, and particularly less-than-supervised deep learning techniques, this had led to numerous proposals for model-independent search strategies for so-termed anomalies. In this presentation, we will review these new techniques with an eye on their underlying assumptions and complementarities.
In this talk I present the important exploratory function that a family of models —what I call perspectival models — play across a number of areas in science. First I explain what I mean by 'perspectival models' then I illustrate their role as 'inferential blueprints' that deliver knowledge about what is possible. I draw on my book Perspectival Realism (OUP, 2022) to illustrate the importance of perspectival models in delivering modal knowledge with examples taken from Beyond Standard Model searches in particle physics.