Learning from imbalanced data streams

Learning in nonstationary environments constitutes a major challenge, and this problem becomes orders of magnitude more complex in the presence of class imbalance. Learning from nonstationary and imbalanced data has been studied separately, but many key challenges remain open when the joint problem is considered.

Representative publications: IEEE TNNLS 2020 [pdf] [code], IEEE SSCI 2022 [pdf], ICANN 2018 [pdf] [code]

Active learning from data streams

Apart from data nonstationarity (concept drift) and class imbalance, acquiring ground truth information (e.g., labels in classification tasks) as instances arrive one-by-one can be costly or impossible in some real-time applications. An effective way to deal with limited labelled data is the active learning paradigm.

Representative publications: Neurocomputing 2022 [pdf] [code], IEEE IJCNN 2020 [pdf], IEEE SSCI 2022 [pdf] [code]

Unsupervised learning (anomaly detection) from data streams

The generation of vast amounts of streaming data in various domains has become ubiquitous. However, many of these data are unlabeled, making it challenging to identify infrequent events, particularly anomalies. This task becomes even more formidable in nonstationary environments where model performance can deteriorate over time due to concept drift. To address these challenges, we propose autoencoder-based incremental learning and concept drift detection mechanisms.

Representative publications: IEEE IJCNN 2024 [pdf], IEEE IJCNN 2023 [pdf], IEEE SSCI 2023 [pdf]

Machine learning for smart water systems

We look into different problems, such as:

• Domestic water consumption monitoring
• Detection and localisation of water contamination
• Urban water consumption forecasting

Representative publications: IEEE TICPS 2024 [pdf], JHI 2024 [pdf], IEEE SSCI 2025 [pdf], IEEE SSCI 2025 [pdf]

Multiagent reinforcement learning for intrusion detection and response

A serious threat in the current Internet is distributed denial of service (DDoS) attacks, which target the availability of the victim system. They are designed to exhaust a server's resources or congest a network's infrastructure, and therefore renders the victim incapable of providing services to its legitimate users. To address this, a distributed and coordinated defence mechanism is necessary, where many defensive nodes, across different locations cooperate in order to stop or reduce the flood. We propose the use of multiagent reinforcement learning to address the problem.

Representative publications: EAAI 2015 [pdf] (Department's Best Student Paper 2015 Award), Connection Science 2015 [pdf], AAAI / IAAI 2013 [pdf], AAMAS 2016 [pdf] [code], EUMAS 2013 [pdf], PhD thesis [pdf]

Multiagent reinforcement learning for downing rogue drones

The wide adoption and use of unmanned aerial vehicles (UAVs) has created not only opportunities but also threats to the security of sensitive areas. Thus, effective and efficient counter-drone systems are required to protect these areas. This work addresses this issue by developing cooperative multi-agent searching, tracking and jamming techniques using RL to counter the operation of one or multiple rogue drones flying over a sensitive area.

Representative publications: IEEE TMC 2024 [pdf], IEEE SMC 2024 [pdf], ICUAS 2023 [pdf]