Code

A selection of open-source code related to ALPhA publications. These repositories live in the alpha-davidson GitHub organization.

• sparse-tpc-embeddings

  Sparse ResNet and PointNet-style encoders built with the Minkowski Engine, with tools to probe their embeddings of TPC events across tasks, experiments, and detectors.

  Paper: How Architecture and Training Affect TPC Representations Across Experiments (arXiv 2026)
  Code: release v2.0.0 · Zenodo: doi:10.5281/zenodo.23926933

  Paper: Sparse Methods for Vector Embeddings of TPC Data (NeurIPS ML4PS 2025)
  Code: release v1.0.0 · Zenodo: doi:10.5281/zenodo.23889820

• attpc_unpaired_trans

  Diffusion probabilistic models for unpaired translation between simulated and experimental AT-TPC events.

  Paper: Unpaired Translation of Point Clouds for Modeling Detector Response (NeurIPS ML4PS 2024)

• SuN_cGAN

  A conditional GAN that unfolds (Ex, Eγ) matrices measured with the SuN detector.

  Paper: Two-dimensional total absorption spectroscopy with conditional generative adversarial networks (NIM A 2024)

• Jet-IQNs

  Implicit quantile networks that model joint distributions for jet emulation, in TensorFlow and PyTorch.

  Paper: Implicit quantile networks for emulation in jet physics (MLST 2024)

• IQNs-for-Jets

  Implicit quantile networks for reconstructing jets from CMS open data.

  Paper: Implicit Quantile Neural Networks for Jet Simulation and Correction (NeurIPS ML4PS 2021)

• falcon-cWGAN

  A fast detector simulator that maps parton-level jets to reconstructed jets with a conditional Wasserstein GAN.

  Paper: Conditional Wasserstein Generative Adversarial Networks for Fast Detector Simulation (CHEP 2021)

• TensorBNN

  Bayesian neural networks trained with Hamiltonian Monte Carlo in TensorFlow.

  Paper: TensorBNN: Bayesian Inference for Neural Networks using TensorFlow (Computer Physics Communications 2022)

ALPhA's research is supported by the National Science Foundation under awards 2311263 and 2012865; see the full list of ALPhA funding. Each repository's README acknowledges the specific funding for that work.