Transfer Learning for Neutrino Scattering: Domain Adaptation with GANs

2025-08-18

AI Summary

Neutrino–nucleus interaction simulations suffer from scarce real data and poor generalization across target nuclei (e.g., carbon → argon) and particle types (neutrino → antineutrino), hindering high-fidelity event generation. Method: This work introduces the first integration of generative adversarial networks (GANs) with domain-adaptive transfer learning to enable cross-scenario event generation at the lepton kinematic level. A GAN is pre-trained on synthetic neutrino–carbon charged-current scattering data, then fine-tuned on small-scale samples (10⁴–10⁵ events) for neutrino–argon and antineutrino–carbon scattering tasks, incorporating domain alignment to mitigate model shift. Contribution/Results: The transferred models significantly outperform from-scratch training, achieving distributional fidelity and physical consistency comparable to high-fidelity simulations—while drastically reducing reliance on large-scale, labeled Monte Carlo data. This establishes a scalable, low-overhead paradigm for high-precision neutrino event generation under data scarcity.

Technology Category

• Computer Vision: Generative Adversarial Networks (GANs) for Vision
• Machine Learning: Transfer, Domain Adaptation, Multi-Task Learning
• Natural Language Processing: Generation

Application Category

• Web Mining and Content Analysis: Web data generation and simulation
• Social Networks and Social Media: Generative AI / large language models and their impact on social systems
• Economics, Online Markets and Human Computation: Economic ramifications for generative AI infrastructure and applications

Abstract

We utilize transfer learning to extrapolate the physics knowledge encoded in a Generative Adversarial Network (GAN) model trained on synthetic charged-current (CC) neutrino-carbon inclusive scattering data. This base model is adapted to generate CC inclusive scattering events (lepton kinematics only) for neutrino-argon and antineutrino-carbon interactions. Furthermore, we assess the effectiveness of transfer learning in re-optimizing a custom model when new data comes from a different neutrino-nucleus interaction model. Our results demonstrate that transfer learning significantly outperforms training generative models from scratch. To study this, we consider two training data sets: one with 10,000 and another with 100,000 events. The models obtained via transfer learning perform well even with smaller training data. The proposed method provides a promising approach for constructing neutrino scattering event generators in scenarios where experimental data is sparse.

Problem

• Adapt GAN model for neutrino-argon and antineutrino-carbon scattering
• Assess transfer learning effectiveness across different interaction models
• Improve event generation with sparse experimental data

Innovation

• Transfer learning for neutrino scattering adaptation
• GANs generate events for different interactions
• Optimizes models with sparse experimental data