CERN and Mistral launch the Flare physics foundation model project
CERN and Mistral have signed a letter of intent for Flare, a foundation model intended to support particle-physics analysis. According to TechNews, the early-stage project plans to train on billions of simulated proton collisions for collision reconstruction, particle identification, and anomaly detection, but no timeline or budget has been announced. CERN and Mistral said the model’s accuracy relative to specialized methods still requires experimental validation.
Flare explores a single foundation model spanning multiple particle-physics analysis steps as CERN reports that its LHC data could grow substantially after the planned 2030 upgrade.
歐洲核子研究組織（CERN）與法國人工智慧公司 Mistral 宣布展開合作，計劃共同開發一款專為科學研究設計的基礎模型，以應對粒子物理領域日益龐大的資料分析需求。雙方已簽署合作意向書，並將首個專案命名為「Flare」，目標是打造出碰撞物理領域的首個基礎模型──「大型物理模型」（Large Physics Model）。
整合分析流程，應對大量數據挑戰
這項計畫的核心在於利用大型強子對撞機（LHC）累積的大量資料來訓練 AI。據悉，Flare 將以數十億次模擬的質子碰撞資料為基礎，學習如何重建碰撞事件、辨識粒子，並協助偵測異常訊號。CERN 目前已擁有約 1 艾位元組（Exabyte）的 LHC 實驗資料，而隨著 LHC 預計於 2030 年進行升級，未來的碰撞次數可能增加十倍，資料量將面臨大幅膨脹。