papers
papers in reversed chronological order.
-
Panda Diplomacy: Foundation model pre-training across particle imaging detectors for high energy and nuclear physics
We show that the same model architecture and self-supervised learning recipe can learn useful particle physics without any labels across three different particle detectors (LArTPC, collider TPC, and water Cherenkov). With 1,000 labeled images Panda V2 acheives or gets close to SOTA performance on a variety of downstream tasks, notably beating Panda V1 and FM4NPP.arXiv, 2026 -
Panda: Self-distillation of reusable sensor-level representations for high energy physics
We successfuly apply DINO-like self-distillation to unlabeled 3D images of particle interactions in a LArTPC. We show the model learns to cleanly separate particles and interaction types in latent space, which reduced the number of labeled images required to reach the state-of-the-art for semantic segmentation by 1,000x.arXiv, 2025 -
10 Million Particle Events: Enabling Foundation Models for Sparse 3D Inverse Problems
We propose a new dataset containing 10 million images of particle interactions with realistic detector response and multiple modalities.NeurIPS 2025 AI4Science Workshop (Spotlight), 2025 -
Particle trajectory representation learning with masked point modeling
We propose a masked point modeling approach for learning representations of particle trajectories from unlabeled LArTPC images. PoLAr-MAE learns to separate tracks from showers almost perfectly without any labels.Machine Learning: Science and Technology, 2025 -
The Optical Two- and Three-dimensional Fundamental Plane Correlations for Nearly 180 Gamma-Ray Burst Afterglows with Swift/UVOT, RATIR, and the Subaru Telescope
The Astrophysical Journal Supplement Series, Jul 2022