main concepts scientists sections
SimplePopularAdvanced

Machine Learns to See the Hidden ✎ express

Original: "An Optimal Observable Machine for reinterpretable measurements in high-energy physics"
· Torben Mohr, Alejandro Quiroga Triviño, Fabian Riemer, Artur Monsch, Matteo Defranchis, Joscha Knolle, Ankita Mehta, Jan Kieseler, Markus Klute
arXiv:2601.08813 · 2026-01-13 · arXiv non-exclusive · independent analysis · 1 min
HEP Phenomenology HEP Experiment
A neural network finds the best way to measure an elusive signal on its own.

When a top quark is born in a particle detector, it immediately decays into a shower of other particles. Catching a hint of new physics in that shower is like hearing a whisper at a rock concert. You need not just a good microphone, but the ability to subtract the roar of the amplifiers and the hum of the crowd.

The new method, called Optimal Observable Machine (OOM), teaches a neural network to find the observable that changes most strongly when the parameter of interest varies. The network knows in advance how the detector distorts the signal and what systematic uncertainties are possible. It's like training your ear to pick out a specific sound while knowing the acoustics of the hall and the noise of the ventilation.

The method was tested on measuring a recently discovered excess of events in top quark pair production. Such an excess could indicate a new particle, but to study it, one must separate signal from background very precisely.

The observable created by the neural network proved more sensitive than standard approaches and remained reusable for future analyses. This means that data collected today can be re-examined tomorrow with new hypotheses — without re-simulating the detector.

The top quark is the heaviest known elementary particle, and its behavior is especially sensitive to unknown forces. The new tool will help us hear even the faintest whisper of new physics in the noise of the Large Hadron Collider.

Whole graph →

The top quark lives for only about $10^{-25}$ seconds — shorter than the time it takes light to cross a proton.

Scientists
Daniel GottesmanDavid DiVincenzoGilles BrassardIgnacio CiracJohn PreskillPeter Zoller
Concepts
observable observation maximum likelihood observer maximum likelihood estimation prediction error Reinforcement Learning robust machine learning generalization orbital angular momentum optically detected magnetic resonance parameter estimation machine learned interatomic potential least squares automated analysis koopman operator neural network regression parameter probabilistic model optimization