Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Network architecture search of X-ray based scientific applications

Adarsha Balaji, Ramyad Hadidi, Gregory Kollmer, Mohammed E. Fouda, Prasanna Balaprakash

TL;DR

The work addresses the computational bottlenecks of Bragg peak detection and ptychographic reconstruction in X-ray diffraction microscopy by automating neural network design through NAS and HPS. Using the Deephyper AutoML framework, the authors optimize both BraggNN and PtychoNN across architecture variants and hyperparameters, achieving substantial reductions in model size while improving accuracy. Multi-objective optimization yields Pareto-optimal solutions that balance accuracy, confidence, and size, with hardware demonstrations on the Orin-AGX edge platform showing meaningful gains in latency and energy efficiency. Overall, automated architecture and hyperparameter search enable faster, more energy-efficient edge inference for high-resolution materials imaging tasks, advancing real-time scientific analysis at the edge.

Abstract

X-ray and electron diffraction-based microscopy use bragg peak detection and ptychography to perform 3-D imaging at an atomic resolution. Typically, these techniques are implemented using computationally complex tasks such as a Psuedo-Voigt function or solving a complex inverse problem. Recently, the use of deep neural networks has improved the existing state-of-the-art approaches. However, the design and development of the neural network models depends on time and labor intensive tuning of the model by application experts. To that end, we propose a hyperparameter (HPS) and neural architecture search (NAS) approach to automate the design and optimization of the neural network models for model size, energy consumption and throughput. We demonstrate the improved performance of the auto-tuned models when compared to the manually tuned BraggNN and PtychoNN benchmark. We study and demonstrate the importance of the exploring the search space of tunable hyperparameters in enhancing the performance of bragg peak detection and ptychographic reconstruction. Our NAS and HPS of (1) BraggNN achieves a 31.03\% improvement in bragg peak detection accuracy with a 87.57\% reduction in model size, and (2) PtychoNN achieves a 16.77\% improvement in model accuracy and a 12.82\% reduction in model size when compared to the baseline PtychoNN model. When inferred on the Orin-AGX platform, the optimized Braggnn and Ptychonn models demonstrate a 10.51\% and 9.47\% reduction in inference latency and a 44.18\% and 15.34\% reduction in energy consumption when compared to their respective baselines, when inferred in the Orin-AGX edge platform.

Network architecture search of X-ray based scientific applications

TL;DR

The work addresses the computational bottlenecks of Bragg peak detection and ptychographic reconstruction in X-ray diffraction microscopy by automating neural network design through NAS and HPS. Using the Deephyper AutoML framework, the authors optimize both BraggNN and PtychoNN across architecture variants and hyperparameters, achieving substantial reductions in model size while improving accuracy. Multi-objective optimization yields Pareto-optimal solutions that balance accuracy, confidence, and size, with hardware demonstrations on the Orin-AGX edge platform showing meaningful gains in latency and energy efficiency. Overall, automated architecture and hyperparameter search enable faster, more energy-efficient edge inference for high-resolution materials imaging tasks, advancing real-time scientific analysis at the edge.

Abstract

X-ray and electron diffraction-based microscopy use bragg peak detection and ptychography to perform 3-D imaging at an atomic resolution. Typically, these techniques are implemented using computationally complex tasks such as a Psuedo-Voigt function or solving a complex inverse problem. Recently, the use of deep neural networks has improved the existing state-of-the-art approaches. However, the design and development of the neural network models depends on time and labor intensive tuning of the model by application experts. To that end, we propose a hyperparameter (HPS) and neural architecture search (NAS) approach to automate the design and optimization of the neural network models for model size, energy consumption and throughput. We demonstrate the improved performance of the auto-tuned models when compared to the manually tuned BraggNN and PtychoNN benchmark. We study and demonstrate the importance of the exploring the search space of tunable hyperparameters in enhancing the performance of bragg peak detection and ptychographic reconstruction. Our NAS and HPS of (1) BraggNN achieves a 31.03\% improvement in bragg peak detection accuracy with a 87.57\% reduction in model size, and (2) PtychoNN achieves a 16.77\% improvement in model accuracy and a 12.82\% reduction in model size when compared to the baseline PtychoNN model. When inferred on the Orin-AGX platform, the optimized Braggnn and Ptychonn models demonstrate a 10.51\% and 9.47\% reduction in inference latency and a 44.18\% and 15.34\% reduction in energy consumption when compared to their respective baselines, when inferred in the Orin-AGX edge platform.
Paper Structure (26 sections, 2 equations, 6 figures, 2 tables)

This paper contains 26 sections, 2 equations, 6 figures, 2 tables.

Table of Contents

  1. Introduction
  2. Background
  3. BraggNN
  4. PtychoNN
  5. Search Methods
  6. Methodology
  7. Neural Architecture Search
  8. BraggNN Architecture Search Space
  9. PtychoNN Architecture Search Space
  10. Hyper-parameter Tuning
  11. Model Optimization
  12. Search Space
  13. BraggNN
  14. PtychoNN
  15. Experimental Setup
  16. ...and 11 more sections

Figures (6)

  • Figure 1: Neural Network models used for X-ray scientific applications; (a)BraggNN and (b) PtychoNN models.
  • Figure 2: BraggNN: Deephyper vs Optuna - Convergence speed.
  • Figure 3: Proposed neural architecture search of the PtychoNN model.
  • Figure 4: BraggNN Model Optimization: pareto front of (a) single objective - model accuracy and (b) multi-objective optimization, respectively
  • Figure 5: PtychoNN Model: Pareto front of (a) single objective - model accuracy and (b) multi-objective optimization, respectively
  • ...and 1 more figures