NTHU seal NTHU · DEPARTMENT OF PHYSICS Accelerated Intelligence & Particle Physics Lab
RESEARCH

From detector to discovery

Our work spans physics analysis, machine-learning algorithms, and computing infrastructure — every layer in service of one goal: finding new physics in LHC data.

01 Beyond-Standard-Model searches 02 Fast ML 03 ML Reconstruction 04 Foundation models 05 Inference as a Service
01

Beyond-Standard-Model searches

ATLAS muon endcap wheel The ATLAS detector · CERN

Exploring new physics through innovative searches for rare and challenging collider signatures.

SELECTED RESULTS Higgs → scalars → 4b/6b — Phys. Rev. D 112 (2025) 072005 Top-associated pseudoscalar → bb̄ — Eur. Phys. J. C 85 (2025) 886 Higgs → aa → 4b (merged) — Phys. Rev. D 102 (2020) 112006
02

Fast ML and real-time AI triggers

ATLAS collision animation ATLAS collision animation (1:27-1:29) · Source

Developing fast machine-learning models and AI triggers for real-time decisions.

SELECTED RESULTS Level-0 MDT trigger FPGA algorithms for the HL-LHC (2018–2023) Event Filter tracking (2023–) b-jet trigger Run 4 R and D (2025–)
03

ML-based Reconstruction

ATLAS inner detector services ATLAS inner detector services · CERN

Using AI to reconstruct complex collision events with greater speed and precision.

SELECTED RESULTS LSH-based Efficient Point Transformer — NeurIPS 2025 ML4PS Workshop · arXiv:2510.07594 DeXTer: Deep Sets for low-pT X → bb̄ ID — ATL-PHYS-PUB-2022-042
04

Foundation models for science

EveNet foundation-model architecture for classification, reconstruction, segmentation, and generation tasks EveNet model architecture · Click to enlarge

Building general-purpose models to understand and analyze diverse scientific data.

SELECTED RESULTS EveNet: A Foundation Model for Particle Collision Data Analysis — arXiv:2601.17126
05

Inference as a Service

SuperSONIC SuperSONIC · Open-source inference infrastructure

Providing scalable, efficient ML inference infrastructure for the scientific community.

SELECTED RESULTS SuperSONIC: Cloud-Native Infrastructure for ML Inferencing — PEARC25 AthenaTriton — Proceedings for CHEP 2025, ATL-SOFT-PROC-2025-026 Track Reconstruction as a Service — JINST 20 (2025) P06002
Want to work on one of these? Join the lab