# STS-AT: A Structured Tensor Flow Adversarial Training Framework for Robust Intrusion Detection

## TL;DR

The paper introduces STS-AT, a new intrusion detection framework that achieves high accuracy and robustness against adversarial attacks.

## Contribution

The novel STS-AT framework combines structured tensor encoding with multi-strategy adversarial training for robust intrusion detection.

## Key findings

- STS-AT achieves 99.6% accuracy in normal traffic classification on the CICIDS2017 dataset.
- Under adversarial attacks, defense accuracy remains above 96.8% after multi-strategy adversarial training.
- Training time is reduced by 67.6% compared to sequential training methods.

## Abstract

What are the main findings?
We propose a Structured Tensor Streaming with Adversarial Training (STS-AT) framework that achieves over 99% accuracy in normal traffic classification and maintains over 96.8% accuracy under various adversarial attacks (FGSM, PGD, and DeepFool).The multi-strategy adversarial training method significantly enhances model robustness, raising defense accuracy from as low as 24.4% (under FGSM attack on undefended models) to above 97.1%.

We propose a Structured Tensor Streaming with Adversarial Training (STS-AT) framework that achieves over 99% accuracy in normal traffic classification and maintains over 96.8% accuracy under various adversarial attacks (FGSM, PGD, and DeepFool).

The multi-strategy adversarial training method significantly enhances model robustness, raising defense accuracy from as low as 24.4% (under FGSM attack on undefended models) to above 97.1%.

What are the implications of the main findings?
We provide a novel methodology that integrates structured raw traffic representation with efficient adversarial training, offering a systematic solution to the dual challenges of feature information loss and model vulnerability in intrusion detection.The efficient training strategy reduces total training time by approximately 67.6% compared to sequential training methods, demonstrating strong potential for practical deployment in real-world network environments.

We provide a novel methodology that integrates structured raw traffic representation with efficient adversarial training, offering a systematic solution to the dual challenges of feature information loss and model vulnerability in intrusion detection.

The efficient training strategy reduces total training time by approximately 67.6% compared to sequential training methods, demonstrating strong potential for practical deployment in real-world network environments.

Network intrusion detection is a key technology for ensuring cybersecurity. However, current methods face two major challenges: reliance on manual feature engineering, which leads to the loss of discriminative information, and the vulnerability of deep learning models to adversarial sample attacks. To address these issues, this paper proposes STS-AT, a novel network intrusion detection method that integrates structured tensors with adversarial training. The method consists of three core components: first, structured tensor encoding, which fully converts raw hexadecimal traffic into a numerical representation; second, a hierarchical deep learning model that combines CNN and LSTM networks to simultaneously learn spatial and temporal features of the traffic; third, a multi-strategy adversarial training method that enhances model robustness by adaptively adjusting the mix of adversarial samples in different training phases. Experiments on the CICIDS2017 dataset show that the proposed method achieves an accuracy of 99.6% in normal traffic classification, significantly outperforming classical machine learning baselines such as Random Forest (93.1%) and Support Vector Machine (84.7%). Crucially, under various adversarial attacks (FGSM, PGD, and DeepFool), the accuracy of an undefended model drops to as low as 24.4%, whereas after multi-strategy adversarial training, the defense accuracy rises above 96.8%. Meanwhile, the total training time is reduced by approximately 67.6%. These results verify that structured tensor encoding effectively preserves original traffic information, the hierarchical model achieves comprehensive feature learning, and multi-strategy adversarial training significantly improves training efficiency while ensuring robust defense effectiveness.