Search NASA⌕ Search
Home/Archive/DOE OSTI

DOE OSTI · 3025679

Variable rate neural compression for sparse detector data

Huang, Yi [Brookhaven National Laboratory (BNL), Upton, NY (United States)] (ORCID:0000000270247469)·Go, Yeonju [Brookhaven National Laboratory (BNL), Upton, NY (United States)]·Huang, Jin [Brookhaven National Laboratory (BNL), Upton, NY (United States)]·Li, Shuhang [Columbia Univ., New York, NY (United States)]·Luo, Xihaier [Brookhaven National Laboratory (BNL), Upton, NY (United States)] (ORCID:000000027621940X)·Marshall, Thomas [Univ. of California, Los Angeles, CA (United States)]·Osborn, Joseph D. [Brookhaven National Laboratory (BNL), Upton, NY (United States)]·Pinkenburg, Christopher [Brookhaven National Laboratory (BNL), Upton, NY (United States)]·Ren, Yihui [Brookhaven National Laboratory (BNL), Upton, NY (United States)]·Shulga, Evgeny [Brookhaven National Laboratory (BNL), Upton, NY (United States); Stony Brook Univ., NY (United States)]·Yoo, Shinjae [Brookhaven National Laboratory (BNL), Upton, NY (United States)]·Yoon, Byung-Jun [Brookhaven National Laboratory (BNL), Upton, NY (United States); Texas A & M Univ., College Station, TX (United States)]
View original source ↗Read source document ↗Export metadata ↓

Abstract

Particle colliders produce data at extraordinary rates, posing major challenges for transmission and storage. High-throughput compression algorithms are therefore essential. In the sPHENIX experiment taking data at the Relativistic Heavy Ion Collider, a time projection chamber records three-dimensional (3D) particle trajectories that are highly sparse, making conventional learning-free lossy compression ineffective. Convolutional neural networks have surpassed traditional methods in compression ratio and accuracy. However, they fail to exploit sparsity for efficiency. To address these gaps, we present BCAE-VS, a bicephalous convolutional autoencoder with variable compression ratio for sparse data, which adapts compression to input complexity through key-point identification and sparse convolution. BCAE-VS achieves higher accuracy and compression ratios than prior neural approaches while being orders of magnitude smaller. Moreover, its throughput increases with sparsity—a property not observed in other methods. Although it was developed for collider experiments, BCAE-VS readily extends to other sparse data domains, such as light detection and ranging (LiDAR) sensing and 3D microscopy.

Explore related subjects

73 NUCLEAR PHYSICS AND RADIATION PHYSICSautoencoderdata compressiondeep learninghigh-energy and nuclear physicshigh-throughput inferencesparse datasparse neural network

Keep this discovery

Explore connections, maps & timelines

BibTeXRIS

Huang, Yi [Brookhaven National Laboratory (BNL), Upton, NY (United States)] (ORCID:0000000270247469), Go, Yeonju [Brookhaven National Laboratory (BNL), Upton, NY (United States)], Huang, Jin [Brookhaven National Laboratory (BNL), Upton, NY (United States)], Li, Shuhang [Columbia Univ., New York, NY (United States)], Luo, Xihaier [Brookhaven National Laboratory (BNL), Upton, NY (United States)] (ORCID:000000027621940X), Marshall, Thomas [Univ. of California, Los Angeles, CA (United States)], Osborn, Joseph D. [Brookhaven National Laboratory (BNL), Upton, NY (United States)], Pinkenburg, Christopher [Brookhaven National Laboratory (BNL), Upton, NY (United States)], Ren, Yihui [Brookhaven National Laboratory (BNL), Upton, NY (United States)], Shulga, Evgeny [Brookhaven National Laboratory (BNL), Upton, NY (United States); Stony Brook Univ., NY (United States)], Yoo, Shinjae [Brookhaven National Laboratory (BNL), Upton, NY (United States)], Yoon, Byung-Jun [Brookhaven National Laboratory (BNL), Upton, NY (United States); Texas A & M Univ., College Station, TX (United States)]. 2026-02-02. Variable rate neural compression for sparse detector data. https://doi.org/10.1016/j.patter.2025.101452

Cite the original work for its findings. Save a collection to share your selection of sources.

KEEP EXPLORING

Related reports

DOE OSTI2026

Design-informed neutronics assessment of long-lived fission product transmutation in a tokamak fusion reactor blanket

This study presents a neutronics-based assessment of the feasibility and viability of transmuting six major long-lived fission products (LLFPs) from light-water reactors, namely 99 Tc, 129 I, 79 Se, 93 Zr, 126 Sn, and 135 Cs, within the blanket region of a tokamak fusion reactor, using the MIT ARC design as a concrete fusion configuration. Monte Carlo neutronics simulations were performed to evaluate LLFP transmutation and to compare the results with a reference boiling water reactor (BWR). The results indicate that transmutation of all six LLFPs is neutronics-feasible in fusion reactors, with transmutation half-lives significantly shorter than their natural decay half-lives. For elemental targets, transmutation of 135 Cs, 126 Sn, and 93 Zr was found potentially viable, as the net mass transmuted exceeded that achievable in the reference BWR under identical target volume and irradiation time. When isotopically separated targets were considered, transmutation of 126 Sn and 93 Zr appeared potentially viable. A parametric study demonstrated that plasma geometry modifications can enhance local neutron flux, increasing the transmuted 93 Zr mass by approximately 33% and reducing the transmutation half-life from approximately 240 years to 180 years. Repositioning the target and adjusting material layer thickness reduced the transmutation half-life of 93 Zr to 67 years and increased the net mass transmuted by a factor of 50. Furthermore, these results demonstrate that fusion reactors can enable LLFP transmutation beyond the practical limits of thermal fission reactors and highlight the critical role of reactor and blanket design optimization. Engineering and fuel-cycle considerations required for deployment are beyond the scope of this neutronics-focused study.

73 NUCLEAR PHYSICS AND RADIATION PHYSICS↗
DOE OSTI2026

A hybrid Monte Carlo-deterministic second moment method with efficient variance reduction

In this work, we present a hybrid method that combines Monte Carlo with deterministic finite element methods to solve a linear Boltzmann transport equation. Our hybrid method runs orders of magnitude faster than Monte Carlo, without sacrificing accuracy, for a proxy problem from radiative transfer that contains both optically-thick and optically-thin material. We believe that this is the first demonstration of a hybrid Second Moment Method in more than one spatial dimension, the first to consider more than one material, and the first to use variance reduction. Our variance reduction approach arises from an asymptotic analysis in which we show that the magnitude of the scattering source grows without bound. We transform the problem to compute the deviation of the radiation intensity from isotropy. The magnitude of the source in the transformed problem is bounded, and the quality of the hybrid method solution is dramatically improved by a substantial reduction in the variance.

73 NUCLEAR PHYSICS AND RADIATION PHYSICS↗