Computing & Intelligence

IHPC focuses on developing knowledge-driven AI modelling frameworks with inputs from both domain knowledge and data. This is accomplished by leveraging on multi-type models such as physics-based, data-driven, and simulation, leading towards accurate, robust, generalisable, and explainable solutions. 

The main research areas include: 

Physics-based AI
To couple AI with physics and/or other types of scientific knowledge, and/or modelling and simulation outputs, such to leverage their complementary strengths and improve performance of target tasks.

Research Topics: 
To make use of different physics knowledge for: 

• Physics-based Data Augmentation: to augment & enrich the data in both quantity and quality.
• Physics-based Representation Learning: to learn features with physical meanings from original data.
• Physics-based Modelling: to build ML models with physics guidance/ incorporation. 

• Data efficiency: to achieve efficient learning from a small amount of labelled data or even no labelled data. We devote to explore on semi-supervised learning/weekly supervised learning, self-supervised learning, transfer learning and unsupervised learning to reduce the data annotation requirement.
• Computation efficiency: to reduce the computation cost of deep AI models, especially on edge devices. We focus on constraint-aware model compression and energy-efficient learning to compress heavy deep model into a lighter one to meet the requirement with constraints on memory and power consumption.

• Learning with Multimodal Representations: to learn the model using multimodal inputs.
• Learning with Auxiliary Modality: to train the model with auxiliary modality which is available during training but not available to testing.
• Learning with Domain Knowledge: to learn the model using domain knowledge in the form of graph and physical rules.

• Auto data pre-processing and feature engineering: to automate the data cleaning, feature generation, and feature selection for data sources of domain specific applications.
• Auto modelling: machine learning (ML) algorithms selection and hyperparameter optimisation; to automate ML pipeline creation and evaluation; to automate multi-model ensemble; to automate modelling with constraints such as time, resource, host device, etc.
• Lifelong learning: continuous monitoring of data characteristics and model performance; continuous knowledge/experience accumulation through past learning; strategy in triggering auto feature engineering and/or auto-modelling.

• Code optimisation for parallel & heterogeneous computing  with modern hardware architectures. 
  
• Edge AI computing: to enable ubiquitous AI deployment catering to demanding scenarios and constraints.
• Quantum computing: to understand quantum computing and quantum-inspired computing paradigms, and apply quantum computing to real-world applications.

• Distributed computation & system for efficient parallel data processing and adaptive resource provisioning on distributed system and framework.
• Distributed learning and interoperability on complex data flow, efficient analytics workflow, flexible model deployment on edge, on-premises and cloud platforms.
• Trusted computing & collaboration to technology convergence of Blockchain, secure multiparty computation and federated learning for trustworthy ecosystem.