TECHNOLOGY & INNOVATION Strategic AI Collaborations through Research Translation @ ST Engineering We earned international recognition for our research through presentations at top global conferences. Highlights include innovations in small model cooperation at EMNLP (Miami, U.S.), traffic optimisation solutions at ICITT (Florence, Italy), and leadership in multi-agent systems at CoCoMARL (Amherst, U.S.). Building on this momentum, we further strengthened our research ecosystem by conferring new Distinguished Professorships under the ST Engineering @ Research Translation programme and establishing strategic collaborations with leading institutions Scaling AI-Driven Development with Coding AI Rollout We successfully rolled out Coding AI across the Group, transforming software development. Now embraced by our engineers, Coding AI utilises advanced AI for code generation, debugging and optimisation. This tool has streamlined workflows, boosted productivity and sped up the time-tomarket for new features. By freeing our engineers to focus on creative problem-solving, we are driving efficiency and elevating software quality across the Group. to advance critical AI research. These partnerships include: • NUS: Agentic workflow management and ethical AI governance. • Duke-NUS: Real-time healthcare monitoring with wearable tech and empathetic AI. • NTU: Persona-based simulations for public safety. • SUTD: Efficient LLMs and AI-driven network optimisation. • Carnegie Mellon: Swarm robotics for autonomous multi-agent systems. These initiatives underscore our commitment to leveraging cuttingedge AI technologies to drive innovation. Empowering Traffic Prediction By leveraging advanced AI urban computing models, we have been able to enhance the prediction of traffic speed and volume, potentially achieving greater accuracy and surpassing state-of-the-art benchmarks — particularly in forecasting speed fluctuations during congestion. The refined model appears to minimise lag in detecting speed drops and predicting recovery patterns, which could improve responsiveness. This progress may contribute to more effective real-time traffic management, optimised urban mobility and reduced travel delays. Innovating to Fortify Cyber Defences AI-related cyberattacks like deepfakes, prompt injection attacks and hacking of AI tools are serious and ever-changing threats. Deepfakes use AI to create fake but realistic videos or audio that can trick people; prompt injection attacks fool AI systems into giving harmful or wrong answers by changing the instructions they receive; while hackers of AI tools such as those used to process text, look for weaknesses to steal sensitive information. To stay ahead, we continuously innovate and develop advanced defences by integrating proactive, realtime threat detection, robust incident response and AI-driven tools that enhance cybersecurity operations. Additionally, we leverage Generative AI to fortify our cyber platforms and drive innovative solutions for comprehensive data protection. EnLighTen, our next-generation AI-powered traffic signal innovation, enables real-time, adaptive guidance to road users and transport authorities, paving the way for more sustainable transport networks. Ramesh Iyer Srikrishna, our senior AI Engineer from the Group Technology Office, presented his research at the Empirical Methods in Natural Language Processing (EMNLP) conference in Miami, U.S. In his presentation, he proposed a collaborative learning framework for training large language models (LLMs) more efficiently. This framework aims to reduce computational costs while maintaining the performance of LLMs, aligning with our commitment to research and development in the areas of Generative AI and efficient LLMs. 24 ST ENGINEERING | ANNUAL REPORT 2024
RkJQdWJsaXNoZXIy ODIwNTc=