Projects
From EEG diffusion models to reinforcement learning environments and training-free model merging, our work spans neuroscience, machine learning, and agentic AI.
EEGDM — EEG Representation Learning via Generative Diffusion Model
ActiveA self-supervised diffusion model for learning rich EEG signal representations, featuring a specialized State-Space Model architecture and latent fusion transformer — applied to downstream tasks such as seizure detection.
View on GitHubGENII — EEG & Brain Source Imaging on Cloud with AI
ActiveAn open-source platform democratising EEG and brain source imaging for epilepsy diagnosis and neurological research, integrating AI and cloud computing to make advanced neuroimaging accessible in clinical practice.
dPickleBall — Reinforcement Learning Environment for Pickleball AI
ActiveA physics-based reinforcement learning environment simulating competitive pickleball gameplay, where AI agents learn to play and compete autonomously — serving as the challenge domain for a national AI, HCI & BCI competition.
NeuroMerging — Training-Free Model Merging via Neuronal Subspace
ActiveA training-free framework for merging multiple fine-tuned models across different tasks, mitigating performance degradation through neuronal subspace decomposition to enable efficient multi-task model composition.
View on GitHubEfficient Agentic AI Adaptation
Lightweight adaptation strategies enabling agentic AI systems to generalise rapidly across diverse tasks and environments.
Safety & Security in Agentic AI
Research into alignment, robustness, and adversarial defence for autonomous AI agents operating in high-stakes settings.
Agentic AI in Business & Enterprise
Deploying multi-agent workflows for real-world enterprise automation, decision support, and intelligent process orchestration.