FOUNDATIONAL AI
The Institute works on basic research in foundations of AI that includes theory and systems as well building statistical models, foundational models, inferencing models, and generative models.
AI Hardware/Software Systems
AI Theory
Responsible and Safe AI
Reasoning AI
Resource-efficient AI
AI Hardware/Software Systems
Goal
To design appropriate hardware and software systems, programming models, and operating systems that can scale to handle future AI workloads and edge AI applications
Domain Leads:
-
- Prof Peh Li Shiuan
-
- Prof He Bingsheng
Key Research Areas:
- Emerging semiconductor devices or materials for AI hardware
- AI computing systems hardware
- AI computing systems software
- Domain-specific AI systems optimizations
AI Theory
Goal
To seek to provide formal, mathematical understanding of phenomena in learning and inference in AI
Domain Leads:
-
- Prof Lee Wee Sun
-
- Assoc Prof Jonathan Scarlett
Key Research Areas:
- Understanding structures in data and model that allow phenomena in generation of realistic data, in-context and few-shot learning, out-of-distribution robustness, privacy in sharing data, causality, etc.
- Providing mathematical understanding of the information theoretic and computational limitations of the above phenomena
Responsible and Safe AI
Goal
To enhance trust in AI by making it safe and ensure security of AI systems through the development of techniques that enable trusted AI models in terms of robustness, fairness, explainability, fairness, veracity, and accountability. It is critical to manage risks while unlocking the potential of AI. Alignment to human values and safety standards and law is therefore critical.
Domain Leads:
-
- Prof Chua Tat Seng
-
- Prof Xiao Xiaokui
Key Research Areas:
- Enhancing Trust in AI Systems
- Ensuring Safety and Security in AI Systems
- Developing Privacy-Preserving AI Technologies
- Mitigating Biases in AI
Reasoning AI
Goal
To design frameworks and algorithms for reasoning and planning to solve commonsense problems and complex open world problems. Developing AI models that can perform multiple tasks in multiple domains will be particularly challenging. They will need to do planning when there is new data under uncertainty.
Domain Leads:
-
- Prof Lee Wee Sun
-
- Prof Ng Hwee Tou
Key Research Areas:
- Single agent reasoning and planning
- Understand role of foundation models in reasoning
- Mode of representation for reasoning and planning
- Optimal policy, human-like behavior, and fairness of multi-agent
Resource-efficient AI
Goal
To ensure efficiency in data requirements, use of computational resources, energy consumption and learning processes while building sustainable and scalable AI systems without compromising or even effectively improving performance. Reducing data requirements and energy consumption during training and inferencing will be critical. That can also allow for edge deployment of AI Models.
Domain Leads:
-
- Prof Ng See Kiong
-
- Assoc Prof Bryan Low
Key Research Areas:
- Efficient Data: reduce the need for data and infusing prior knowledge
- Efficient Model: parameters-efficient models and multi-modality
- Efficient Strategy: automation, collaboration, and reusability
- Efficient Compute: AI accelerators with hardware-software co-design
© National University of Singapore. All Rights Reserved.