Daily Summary for 2026-10-20
daily
Transcript
Introduction to the show: ident: High Energy Physics: Experiment. The day in high-energy experiment: detectors, searches and limits.
Kai: It's the twentieth of October, twenty twenty-six, and this is the day's research.
Mira: One new paper came out today.
Kai: I'm Kai.
Mira: And I'm Mira, here with Lev.
Kai: We'll take the day in one pass, then pull out the paper we're staying with.
Mira: Start us off with Machine Can Automatically Discover Parametric Functions to Model HEP Data.
The day's research: Kai: The focus today is on automating model finding for experimental data.
Mira: Manually guessing functions for binned data in high energy physics takes a lot of time.
Lev: Machine learning could replace the tedious trial and error process then?
Kai: It suggests machine learning can automate that fitting process.
Mira: What specific types of mathematical models are we testing with this approach?
Lev: We are looking at how machine learning can find those models.
Kai: The core idea involves symbolic regression.
Mira: That technique lets a machine search through vast spaces of possible functions.
Lev: It does this without needing prior knowledge about what kind of function we should be looking for.
Kai: This is put into practice with the SymbolFit package.
Mira: SymbolFit combines this search with uncertainty modeling.
Lev: That modeling is specifically tailored for HEP analysis needs.
Kai: This automated search used five hundred sixty independent seeded runs with seven fit configurations.
Mira: The chi two over degrees of freedom was held around one for those fits.
Lev: One hundred eleven runs successfully rediscovered the dijet and UA2 functions from published searches.
Kai: That means the machine found known good models in those specific dijet spectrum analyses.
Mira: This suggests the automated approach can speed up finding appropriate functional forms for complex data.
Lev: The paper Machine Can Automatically Discover Parametric Functions to Model HEP Data shows how a machine can automatically find the best mathematical function to describe complex physics data without needing prior knowledge of what that function should look like.
Kai: SymbolFit uses symbolic regression to search for suitable functions and demonstrated success on real particle physics data from CMS and ATLAS experiments.
Mira: The tool successfully rediscovered previously used functions in dijet spectrum analyses.
Lev: Agreed, this shows the method's capability with established functional forms.
Lucky paper: 2607.19750: Kai: So, we're looking at how this paper Machine Can Automatically Discover Parametric Functions to Model HEP Data actually tackles the manual process.
Mira: The introduction points out that modeling binned data with smooth functions is everywhere in high energy physics, but deriving those functions from first principles is hard.
Lev: Most practical binned distributions don't have a true underlying function, so you can only find an adequate functional form empirically.
Kai: The standard practice involves guessing a form by intuition and then fitting it to data while checking the residuals.
Mira: That iterative process of fine-tuning the form and refitting is what the authors call empirical, and it can be slow for analysis progress.
Lev: The paper says that final candidate functions often get stuck in a small corner of function space because they are too fine-tuned to that specific analysis.
Kai: So, this machine learning method, symbolic regression SR, performs a data-driven search over the function space instead of relying on prior knowledge.
Mira: The functional form becomes a trainable parameter that changes during the fit to minimize the residual loss.
Lev: The paper introduces SymbolFit which streamlines this into three steps: first function search, second re-optimization and uncertainty estimation, and third evaluation.
Kai: That means PySR searches for functions over a function space using a multi-population evolutionary algorithm without knowing what the form should look like beforehand.
Mira: Then there's the re-optimization step where LMFIT interfaces with it to compute uncertainties via the covariance matrix for each returned function.
Lev: The authors show that this method not only finds functions that model spectra but also discovers wide ranges of different functions, including the dijet and UA2 functions.
Kai: They test this by using two public datasets from HEPData: CMS Run two observed dijet mass spectra at sqrt s = thirteen TeV and ATLAS Run two observations.
Mira: Specifically, they use Figure five from Ref one and Figure 3a from Ref two, where the dijet function in Eq. one was used to model the spectra in the published results.
Lev: The paper also mentions that a search typically returns tens of functions stochastically drawn from a large function space, which can be repeated with different random seeds.
More episodes
- Daily Summary for 2026-11-19
- 2512.21290-Model-independent ZH production cross section at FCC-ee
- 2610.21982-First Observation of the Radiative Decay D 0 to gamma omega
- The day in conversation for 2026-11-19
- Daily Summary for 2026-11-19
- Daily Summary for 2026-11-18
- 2512.08065-Searches for Light Dark Matter and Evidence of Coherent Elastic Neutrino-Nucleus Scattering of Solar Neutrinos with the LUX-ZEPLIN (LZ) Experiment
- Daily Summary for 2026-11-18
- 2508.18022-Precision Measurements of the Electroweak Mixing Angle in the Region of the Z pole at the Large Hadron Collider
- 2512.23992-Prospect for measurement of CP-violating parameters of B s 0 to phi gamma at the Tera Z factory