Shaping the Future of Responsible AI: CCDS at the AI Symposium 2025

At the heart of the symposium was a keynote by Professor Luke Ong, Vice President (AI & Digital Economy) and Dean of CCDS, who presented a bold, theoretically grounded vision for Safe Reinforcement Learning. 

“If we want AI to serve society, we must design systems that are not only intelligent, but also interpretable and accountable. The future of AI lies not just in scale, but in responsibility; and that requires deep collaboration across disciplines and borders,” said Prof Ong. 

 

 

Reinforcement Learning: An AI Safety Perspective 
 
Prof Ong’s talk introduced a novel approach that uses Linear Temporal Logic (LTL) to specify both objectives and safety constraints in reinforcement learning; transforming AI decision-making into a constrained optimisation problem grounded in formal logic. His research tackles long-standing issues like reward hacking and lack of interpretability, offering a clear path toward safe-by-design reinforcement learning.

• Automated red teaming to uncover vulnerabilities 

• Machine unlearning to remove unsafe capabilities 

• Mechanistic interpretability for understanding LLMs 

• Formal verification frameworks to ensure guaranteed safe AI 

This keynote underscored CCDS’s unique ability to bridge deep theory, practical implementation, and ethical foresight; a trifecta essential for the next phase of AI development. 

 

Keynotes from CCDS Faculty 
 
AI for a Sustainable Digital Future 
 
Prof Wen Yonggang, President’s Chair of CCDS, Associate Provost (Graduate Education), and Dean (Graduate College) addressed the environmental implications of deep learning. He noted that while AI contributes to environmental burden, it also holds promise for sustainable solutions. Prof Wen emphasised the importance of building systems that are not only more efficient, but also environmentally friendly and aligned with global sustainability goals as he urged researchers to rethink how we AI systems are built and used. 

 

Foundation Model Fusion 
 
Prof Tao Dacheng, distinguished University professor at CCDS, delivered a keynote on the evolution of AI into the era of Foundation Models. His team’s research focuses on fusion and alignment of multiple foundation models, allowing accelerated training at a fraction of the traditional cost; achieving comparable performance with greater efficiency. 

 

Human-Like Perception in Machines 
 
Prof Lin Weisi, CCDS Associate Dean (Research) and President’s Chair Professor in computer science,  explored how AI systems can better mimic human perceptual processes, particularly in visual signal interpretation. His team has developed novel AI architectures that improve human-machine interaction and significantly reduce computational load. 

 

Brain-Computer Interfaces (BCI) 

• Enhancing motor skills in stroke patients 

• Improving focus in children with ADHD 

• Decoding speech in individuals who have lost the ability to speak 

• Alleviating anxiety symptoms through signal-based interventions 

 

Generative AI & Computer Vision 
 
Prof Loy Chen Change, President’s Chair Professor in computer science, delivered a keynote that challenged conventional paradigms in computer vision, proposing a generative-first approach that enhances data diversity, supports open-vocabulary reasoning, and unifies perception with synthesis. 

 

Leading Across Domains 
 
Beyond the keynote presentations, CCDS faculty anchored multiple thematic sessions: 
 
AI in Healthcare and Bioinformatics 
 
Prof Jagath Rajapakse presented a suite of explainable and generative tools for: 

• Functional MRI interpretation 

• Drug response prediction using gene–compound data 

• New molecule generation via GenAI
  

He also highlighted CCDS’s involvement in international efforts to build AI-enabled, integrated healthcare systems; alongside researchers from UCL and Hungary. 
 
Graph Science and Quantum AI 
 
Prof Cong Gao explored cohesive subgraph algorithms and graph neural networks, with applications in urban mobility modelling and traffic prediction; even extending to quantum-enhanced methods. 

Uncertainty & Explainability 
 
Prof Daniel Paulin, who also served as one of the programme coordinators for the symposium, introduced a new Bayesian sampling method for uncertainty quantification in AI diagnostics. His work showed significant gains in pneumonia detection accuracy from X-ray images; a critical step toward trustworthy clinical AI.

L-R: Barbara Bálint (HUN-REN), András Benczúr (Institute for Computer Science and Control, Hungary), Daniel Paulin and Luke Ong.

• Trustworthy and sustainable machine learning 

• Security and adversarial robustness 

• AI governance and policy frameworks 

CCDS’s strong faculty presence – across keynote, panels, and research showcases – signalled not just technical excellence, but a clear commitment to leading the global conversation on safe, fair, and impactful AI. 
 
From theoretical breakthroughs to real-world applications, CCDS demonstrated that thought leadership in AI isn't just about being present in global dialogues.

It’s about shaping them.