Abstract:Adversarial training with Normalizing Flow (NF) models is an emerging research area aimed at improving model robustness through adversarial samples. In this study, we focus on applying adversarial training to NF models for gravitational wave parameter estimation. We propose an adaptive epsilon method for Fast Gradient Sign Method (FGSM) adversarial training, which dynamically adjusts perturbation strengths based on gradient magnitudes using logarithmic scaling. Our hybrid architecture, combining ResNet and Inverse Autoregressive Flow, reduces the Negative Log Likelihood (NLL) loss by 47\% under FGSM attacks compared to the baseline model, while maintaining an NLL of 4.2 on clean data (only 5\% higher than the baseline). For perturbation strengths between 0.01 and 0.1, our model achieves an average NLL of 5.8, outperforming both fixed-epsilon (NLL: 6.7) and progressive-epsilon (NLL: 7.2) methods. Under stronger Projected Gradient Descent attacks with perturbation strength of 0.05, our model maintains an NLL of 6.4, demonstrating superior robustness while avoiding catastrophic overfitting.

What problem does this paper attempt to address?

The main problem that this paper attempts to solve is to improve the robustness of gravitational - wave parameter estimation, especially in the face of data perturbations and detector noise. Specifically, the author focuses on how to enhance the robustness of gravitational - wave signal parameter estimation through adversarial training combined with normalizing - flow models. ### Main problems: 1. **Accurate parameter estimation of gravitational - wave signals**: Gravitational - wave signals are usually buried in detector noise, so accurately extracting astrophysical parameters (such as mass, spin, distance, etc.) from noisy data is a key challenge. 2. **Sensitivity of the model to data perturbations**: Traditional parameter estimation methods are fragile in the face of data perturbations, especially in gravitational - wave detection, where the detector noise characteristics change significantly over time. 3. **Balance between computational efficiency and robustness**: A method is required that can process data quickly and maintain high precision under various perturbations. ### Solutions: To address the above challenges, the paper proposes a new method, that is, enhancing the robustness of the normalizing - flow model through adaptive adversarial training. Specific innovation points include: - **Adaptive epsilon method**: Dynamically adjust the perturbation intensity based on the gradient magnitude, and use logarithmic scaling to handle features of different scales. - **Hybrid architecture**: Combine ResNet for feature extraction and Inverse Autoregressive Flow (IAF) for probability modeling to form a hybrid architecture specifically for gravitational - wave analysis. - **Training objective**: Combine the negative log - likelihood losses of clean samples and adversarial samples to ensure that the model can still maintain good performance under adversarial attacks. ### Experimental results: Experiments show that this method exhibits significantly better robustness than traditional methods when facing FGSM attacks of different intensities and stronger PGD attacks, and its performance on clean data is also close to that of the baseline model. ### Summary: By introducing adaptive adversarial training and a hybrid deep - learning architecture, the paper successfully improves the robustness and computational efficiency of gravitational - wave parameter estimation, providing a solid foundation for reliable data analysis in practical applications.

Adaptive Epsilon Adversarial Training for Robust Gravitational Wave Parameter Estimation Using Normalizing Flows

Bag of Tricks for FGSM Adversarial Training

Gravitational-wave parameter estimation with autoregressive neural network flows

Robust inference of gravitational wave source parameters in the presence of noise transients using normalizing flows

Efficient Gravitational Wave Parameter Estimation via Knowledge Distillation: A ResNet1D-IAF Approach

Adaptive-Gravity: A Defense Against Adversarial Samples

Stable Training of Normalizing Flows for High-dimensional Variational Inference

GEAR: A Margin-based Federated Adversarial Training Approach

Advancing Space-Based Gravitational Wave Astronomy: Rapid Parameter Estimation via Normalizing Flows

Efficient parameter inference for gravitational wave signals in the presence of transient noises using temporal and time-spectral fusion normalizing flow

Transfer Learning Adapts to Changing PSD in Gravitational Wave Data

FL-AGN: A Privacy-Enhanced Federated Learning Method Based on Adaptive Gaussian Noise for Resisting Gradient Inference Attacks

Sampling-based Fast Gradient Rescaling Method for Highly Transferable Adversarial Attacks

Robust Single-step Adversarial Training with Regularizer

Neural Sinkhorn Gradient Flow

Towards Robust Training of Neural Networks by Regularizing Adversarial Gradients

ZeroGrad : Mitigating and Explaining Catastrophic Overfitting in FGSM Adversarial Training

Accelerated Bayesian parameter estimation and model selection for gravitational waves with normalizing flows

DualFlow: Generating imperceptible adversarial examples by flow field and normalize flow-based model