Are you one of the authors of this paper? Email us from your institutional or work email address mentioning this article's arXiv ID and we'll verify you and give you edit access to this page.
Are you an author who sees something we missed or got wrong? Write to us — the subject line already carries the paper id. We read every note and correct the write-up.
A particle trap learns to pick out the important stuff on the fly.
The MicroBooNE detector is a large tank of liquid argon that catches particles flying through it. When a particle passes, it leaves a trail of charge and a flash of light. These trails combine into a three-dimensional picture of the event, as if the particle were drawing its path in a transparent cube.
But there are too many such events: hundreds of cosmic muons fly through the detector every second. Recording everything would mean drowning in data. Usually, the selection is done later, on computers, but that is slow and expensive. The new work shows how to teach the detector to select what matters right on the fly, like an experienced chef skimming the foam off a broth without turning off the stove.
The algorithm looks only at the charge left by the particle in the wire planes and decides in fractions of a second: is this an interesting electron from a stopping muon, or just noise?
Such online selection has been tested for the first time on real data from a liquid argon detector. It works fast and accurately, and it can be transferred to future experiments where the data flow will be even more powerful. It's like teaching a mail filter to tell important letters from spam before they hit your inbox.
Now the detector doesn't just see particles: it learns to hear the whisper of rare events amid the roar of cosmic rays.
