Semi-supervised Variational Temporal Convolutional Network for IoT Communication Multi-anomaly Detection

Yifei Jia, Yongliang Cheng, Jun Shi2022 · DOI

TL;DR

This paper introduces SS-VTCN, a novel semi-supervised deep learning model that effectively detects multiple types of anomalies in IoT communication traffic. By combining Temporal Convolutional Networks (TCNs) for sequence modeling and Variational Autoencoders (VAEs) for representation learning, SS-VTCN leverages both labeled and unlabeled data to achieve superior anomaly detection performance compared to existing methods.

What Problem Does This Paper Address?

The proliferation of Internet of Things (IoT) devices has created massive communication networks, but these devices are often insecure, making them vulnerable to attacks and system failures. These vulnerabilities can manifest as various anomalies in IoT communication traffic, which are difficult to detect accurately and in a timely manner. Traditional anomaly detection methods struggle with the sheer volume and complexity of IoT data, especially when labeled data is scarce. Supervised methods require extensive labeled data, which is costly and time-consuming to obtain for IoT communications. Unsupervised methods, while not needing labels, often fail to categorize anomalies effectively. This paper addresses the challenge of detecting multiple types of anomalies in IoT communication data, particularly in scenarios where only a small portion of the data is labeled.

Methods

The core of the proposed solution is the SS-VTCN (Semi-supervised Variational Temporal Convolutional Network). This model ingeniously combines two powerful deep learning architectures:

  • Temporal Convolutional Network (TCN): TCNs are chosen for their ability to effectively capture long-term temporal dependencies in sequential data. Unlike recurrent neural networks (RNNs) like LSTMs, TCNs use causal convolutions and dilated convolutions, allowing them to process sequences in parallel and maintain a larger receptive field without the vanishing gradient problem. This makes them well-suited for analyzing the chronological nature of IoT communication data.
  • Variational Autoencoder (VAE): VAEs are employed as a generative model. They learn to encode input data into a lower-dimensional latent space and then reconstruct the original data from this latent representation. In SS-VTCN, VAE is used to learn robust representations of both labeled and unlabeled data, helping to generalize the model and improve its ability to reconstruct normal patterns.

The SS-VTCN architecture integrates TCN and VAE to perform both anomaly detection and representation learning. A key innovation is a novel semi-supervised loss function that effectively utilizes both labeled and unlabeled data. This loss function comprises two parts: one for the TCN (focused on discriminative tasks with labeled data) and one for the VAE (focused on reconstruction with all data). Additionally, the paper introduces rectification principles to refine the preliminary anomaly predictions, further enhancing accuracy.

Datasets

The paper evaluates SS-VTCN on two datasets:

  • DS2OS-B: This is a real-world dataset sampled from the Distributed Smart Space Orchestration System (DS2OS), which manages various smart devices in a home environment. It contains communication information with a dimension of 11 and is labeled with four classes: normal, DoS attack, malicious, and spying. Notably, it's a relatively balanced dataset with approximately 50% abnormal records.
  • DS2OS-T: This is a common test dataset, separate from DS2OS-B, used to assess the generalization performance of the detection models.

Key Insights and Ideas

  • The Power of Semi-Supervision in IoT: The paper strongly advocates for semi-supervised learning in IoT anomaly detection, demonstrating that it significantly outperforms purely supervised methods when labeled data is scarce.
  • TCN's Superiority for Temporal IoT Data: The experiments highlight TCN's effectiveness in capturing complex temporal patterns in IoT communication, outperforming LSTM-based models in many scenarios.
  • VAE for Robust Representation: VAEs are crucial for learning generalized representations from unlabeled data, enabling the model to better distinguish normal from anomalous patterns.
  • Rectification Principles Matter: The proposed rectification principles are shown to be critical for improving the accuracy of anomaly selection, especially in multi-anomaly detection scenarios.
  • Optimal Labeling Strategy: The research suggests that a moderate amount of labeled data (around 40%) can yield excellent performance, balancing the cost of labeling with detection effectiveness.

Key Results

Extensive experiments on the DS2OS-B and DS2OS-T datasets demonstrate that SS-VTCN consistently outperforms other compared models, including supervised (LSTM, TCN, VLSTM, VTCN) and other semi-supervised (SS-VLSTM, SS-WVTCN) approaches.

  • Overall Performance: SS-VTCN achieves the best average F1 scores across various classes and labeled data rates. For instance, at 40% labeled data, SS-VTCN shows significant improvements over other models.
  • Semi-Supervised Advantage: Semi-supervised models generally show a 10%-50% improvement in F1 score compared to supervised models, underscoring the value of leveraging unlabeled data.
  • TCN vs. LSTM: Models based on TCN consistently perform better than those based on LSTM, indicating TCN's suitability for IoT communication anomaly detection.
  • Impact of Rectification: SS-VTCN with rectification principles performs better than its variant without them (SS-WVTCN), highlighting the importance of the proposed anomaly selection mechanism.

Limitations and Future Work

  • Small Anomaly Detection: The authors acknowledge that the performance for detecting small anomalies could be further improved.
  • Future Directions: The researchers plan to test SS-VTCN on larger datasets and explore ways to enhance its capability for solving more complex anomaly detection problems.

Summary updated

Abstract

The consumer Internet of Things (IoT) have developed in recent years. Mass IoT devices are constructed to build a huge communications network. But these devices are insecure in reality, it means that the communications network are exposed by the attacker. Moreover, the IoT communication network also faces with variety of sudden errors. Therefore, it easily leads to that is vulnerable with the threat of attacker and system failure. The severe situation of IoT communication network motivates the development of new techniques to automatically detect multi-anomaly. In this paper, we propose SS-VTCN, a semi-supervised network for IoT multiple anomaly detection that works well effectively for IoT communication network. SS-VTCN is designed to capture the normal patterns of the IoT traffic data based on the distribution whether it is labeled or not by learning their representations with key techniques such as Variational Autoencoders and Temporal Convolutional Network. This network can use the encode data to predict preliminary result, and reconstruct input data to determine anomalies by the representations. Extensive evaluation experiments based on a benchmark dataset and a real consumer smart home dataset demonstrate that SS-VTCN is more suitable than supervised and unsupervised method with better performance when compared other state-of-art semi-supervised method.
BibTeX
@misc{jia2022semi,
  author = {Yifei Jia and Yongliang Cheng and Jun Shi},
  title = {{Semi-supervised Variational Temporal Convolutional Network for IoT Communication Multi-anomaly Detection}},
  year = {2022},
  doi = {10.1145/3562007.3562019},
}
Figure