The Internet of Things(IoT) paradigm provides persistent sensing and data collection capabilities and is becoming increasingly prevalent across many market sectors. However, most IoT devices emphasize usability and function over security, making them very vulnerable to malicious exploits. This concern is evidenced by the increased use of compromised IoT devices in large scale bot networks (botnets) to launch distributed denial of service(DDoS) attacks against high value targets. Unsecured IoT systems can also provide entry points to private networks, allowing adversaries relatively easy access to valuable resources and services. Indeed, these evolving IoT threat vectors (ranging from brute force attacks to remote code execution exploits) are posing key challenges. Moreover, many traditional security mechanisms are not amenable for deployment on smaller resource-constrained IoT platforms. As a result, researchers have been developing a range of methods for IoT security, with many strategies using advanced machine learning(ML) techniques. Along these lines, this paper presents a novel generative adversarial network(GAN) solution to detect threats from malicious IoT devices both inside and outside a network. This model is trained using both benign IoT traffic and global darknet data and further evaluated in a testbed with real IoT devices and malware threats.

翻译：物联网范式提供了持续的感知和数据收集能力，并正日益渗透到众多市场领域。然而，大多数物联网设备在功能与安全性之间更侧重于易用性，这使得它们极易遭受恶意攻击。这种担忧的佐证是：大型僵尸网络（botnets）中越来越多地利用被攻陷的物联网设备，对高价值目标发起分布式拒绝服务（DDoS）攻击。不安全的物联网系统还可能成为进入私有网络的入口点，使攻击者能够相对轻松地访问有价值资源和服务。事实上，这些不断演变的物联网威胁载体（从暴力破解攻击到远程代码执行漏洞）正在带来严峻挑战。此外，许多传统安全机制并不适用于资源受限的小型物联网平台。为此，研究人员已开发出一系列物联网安全方法，其中许多策略采用了先进的机器学习（ML）技术。基于此，本文提出了一种新颖的生成对抗网络（GAN）解决方案，用于检测网络内外恶意物联网设备带来的威胁。该模型采用良性物联网流量与全球暗网数据联合训练，并在配备真实物联网设备与恶意威胁的测试平台上进行了进一步评估。