← 研踪 PaperTrailarXiv:2611.26291 · v1
Who Moved My $Z$ Boson? Revealing Resonance Features Inside a Neural Network
首次提交 2026-11-19 23:48
版本更新 2026-11-19 23:48
雷达观测 2026-11-20 14:52
摘要
Methods used in mechanistic interpretability have revealed human-aligned concepts inside language and image models. In this paper, we search for physics concepts inside a neural a generative collider physics model trained on real LHC data (ShellFlow). We discover features inside ShellFlow that represent resonances, such as the , , and . The network was never told which muons come from a resonance, it was only trained to generate the kinematics of the final-state particles. We use these features to alter aspects of the boson: we change its mass between and GeV, and removing such a feature from the neural network produces a broader peak. In both cases the rest of the invariant mass spectrum remains mostly unchanged. The feature was built from the gradients of the neural network, and its selectivity was built using pattern recognition methods. A generative particle physics model trained on unlabeled data can be modified internally to change a physical quantity.
作者备注:10 pages (+4 pages of references, +14 pages of supplementary material, 29 total pages), 7 figures in the main text (+14 figures in the supplementary material, total of 21 pages)
观测到的版本
相邻记录可能跳过中间版本;首次发现 v2 及以上也会标为版本更新。
v1 · 2026-11-19 23:48 · 最新观测版本
Who Moved My $Z$ Boson? Revealing Resonance Features Inside a Neural Network
Methods used in mechanistic interpretability have revealed human-aligned concepts inside language and image models. In this paper, we search for physics concepts inside a neural a generative collider physics model trained on real LHC data (ShellFlow). We discover features inside ShellFlow that represent resonances, such as the , , and . The network was never told which muons come from a resonance, it was only trained to generate the kinematics of the final-state particles. We use these features to alter aspects of the boson: we change its mass between and GeV, and removing such a feature from the neural network produces a broader peak. In both cases the rest of the invariant mass spectrum remains mostly unchanged. The feature was built from the gradients of the neural network, and its selectivity was built using pattern recognition methods. A generative particle physics model trained on unlabeled data can be modified internally to change a physical quantity.
查看此版本解读 →查看此版本原文 ↗