A key challenge in reinforcement learning is enabling agents to generalize their experiences, applying knowledge gained in one environment to new and varied contexts. Generalizability is essential for success in real-world applications, where agents must adapt to distribution shifts and contextual variations. In this work, we propose a novel framework that integrates visual domain randomization with multimodal foundation models to improve the robustness and adaptability of reinforcement learning agents. This integration allows agents to learn policies that are resilient to environmental changes and visual discrepancies. We evaluate our method in the MiniGrid benchmark, including the unseen test environment (DistShift1), where it achieves a mean return of 0.85, outperforming the Proximal Policy Optimization baseline (0.32). These results show the effectiveness of our framework in addressing distribution shift and highlight its potential for real-world RL applications.

Federated learning enables distributed clients to train a global model while maintaining privacy and control over their data. Although collaboration can substantially improve the learning process, it also introduces vulnerabilities, as not all participants contribute beneficially. Clients may engage in detrimental activities, such as data poisoning, that compromise the integrity of the global model. Additionally, in a realistic scenario, the quality of the data possessed by these clients is highly heterogeneous, which influences the model training performance. Moreover, malicious clients (aka free riders) intend to obtain the global model without making a real contribution to the training process. Hence, a reliable and fair evaluation of the client contribution is essential to promote diverse client engagement, improve robustness, and address the free-rider problem. This paper proposes a reputation-aware contribution evaluation approach (RBFL) that provides adversarial robustness by tracking reputation over multiple training rounds to ensure that clients consistently contribute positively. We employ CosineGradient as the utility function and Truncated Monte Carlo (TMC) Shapley as the data valuation function. Empirical evaluation demonstrates the effectiveness of our approach in a fair evaluation of clients’ contributions and effective identification of adversarial clients while maintaining a model accuracy of $92 \%$ with adversarial robustness.

The Internet of Things (IoT) is an impending area with applications in numerous fields. The number of IoT devices has seen exponential growth, increasing apprehensions around security. Cyberattacks are of rising concern because of the expanded attack surface of threats that have plagued networks. Adding to that are insecure practices among users who may not know to protect their IoT devices. Therefore, IoT security has become fundamental, especially as IoT devices carry sensitive data. This paper provides a proof of concept of an Intelligent Intrusion Detection System for IoT. We centered our work on a privacy-preserving approach offering a Federated Learning (FL) based solution for intrusions recognition combining network and energy data. Our model has achieved high accuracy while preserving a short running time in multiple FL rounds.

In the burgeoning age of digitalization, the Internet of Things presents a core part of the digital ecosystem. Unfortunately, as the deployment of connected devices is increasing tremendously, so are cyber-attacks. The consequences of cyber-attacks could be devastating as they gain access to sensitive data and even damages critical infrastructures. This urges the development and integration of proactive and intelligent security breach detection mechanisms in different levels of the IoT platforms including the devices themselves. Several empirical observations indicated a change in the energy consumption and network behaviour of compromised devices. Thus, we propose in this paper a machine learning based approach to identify compromised IoT devices using their energy consumption footprint and network traffic. We base our study on real data collected from real experiments using different commercially available IoT devices infected with authentic IoT botnets. Our results show that machine learning algorithms can classify correctly attacks reaching 98.40% precision for Mirai, over 99.91% for Ufonet and respectively 97.63% and 99.93% performance. Overall, our exploratory study is one of the very first of its kind to explore the energy consumption combined with network behavior analysis to detect IoT compromised devices and its outcomes will be a starting point for further research on this topic.