The following themes ARE being studied in my group:

o   AI for Biology (AI4Bio): with emphasis on multiscale foundation models, genome and transcriptome analysis, protein modeling, clinical applications, and scalable multimodal frameworks to address complex biological challenges and enable actionable insights across scales. Of particular interest are:

1)     Multi-scale foundation models for simulating and predicting biological systems across scales

2)     Genome and transcriptome analysis using hierarchical latent spaces to model DNA, RNA, and protein functions

3)     Protein modeling systems for understanding regulatory logic and enabling functional design

4)     Clinical and healthcare applications addressing disease heterogeneity and advancing personalized treatment

5)     Multi-scale biological models connecting molecular mechanisms to organism-level phenomena

6)     Scalable, efficient frameworks integrating multimodal biological data for complex analysis

Representative work: AIDO (AI-Driven Digital Organism)

o   Artificial General Intelligence (AGI): with emphasis on developing world and agent models to simulate real-world reasoning, address embodied and social reasoning gaps, and explore the philosophical foundations of intelligence and agency. Of particular interest are:

1)     World models simulating real-world possibilities for physical, social, and biological reasoning

2)     Agent models incorporating planning, belief systems, goal-oriented behaviour, and environmental interaction

3)     Addressing gaps in embodied reasoning, social dynamics, and strategic decision-making

4)     Philosophical exploration of intelligence and agency through reasoning frameworks

Representative work: Pandora

o   Large Language Models (LLMs): with emphasis on open-source development, cost-efficient scaling, transparency, domain-specific adaptations, and integration of LLMs as components in broader reasoning and action systems. Of particular interest are:

1)     Open-source initiatives like K2, Jais 70B, and Vicuna as collaborative academic efforts

2)     Scaling and optimization strategies to reduce costs, improve speed, and enhance eco-friendliness

3)     Transparency and reproducibility through LLM360, promoting open access and academic engagement

4)     Domain-specific adaptations enhancing contextual reasoning and addressing specialized tasks

5)     Integration of LLMs as components within broader frameworks for world and agent modeling

Representative work: LLM360

The following themes have been studied in my group:

o   Core Machine learning: represented by :

1)     Theory and algorithms for learning time/space varying-coefficient models with evolving structures or sample-specific (personalized) structures

2)     Meta ML and trustworthy ML for generalizable and adversary-robust algorithms

3)     The "standard equation" for ML: building a unifying framework for various ML paradigms via a standardized forms of loss, model, and solver.

4)     Theory and algorithms for learning sparse structured input/output models and multi-task models in ultra high-dimensional space

5)     Nonparametric Bayesian methods, infinite mixture models, algorithms and applications of Bayesian nonparametrics for data mining and object/topic/event tracking in open, evolving possible worlds

6)     Nonparametric graphical models, RKHS embedding and spectrum algorithm for general graph models

7)     Distributed and online algorithms for optimization, approximate inference, and Monte Carlo sampling on large-scale data and models

o   System Architecture and Strategies for Large Scale ML: with emphasize on developing general purpose systems for machine learning on massive data with massive model on industrial-scale multicore and distributed systems. Of particular interest are:

1)     Design and implementation of representations and systems for composable ML parallelism

2)     Global and local protocols for adaptive scheduling in multi-tenant multi-job distributed ML

3)     Theoretical analysis of distributed ML system behaviors

4)     Automated model learning and tuning via neural architectural search (NAS) and hyperparameter optimization (HPO)

o   Healthcare and Medical Applications: with emphasis on developing algorithms and solutions that address problems of practical clinical, medical, and biological concerns. Of particular interest are:

1)     Robot radiologist: reasoning on rediological images, clinical case report generation, medical training image generation

2)     EHR-based patient modeling and prediction, ICD coding

3)     Sample specific models for panomic-microenvironment interactions in cancer development or cell differentiation via joint analysis of genomic, proteomic, cytogenetic and pathway signaling data

4)     Statistical inference on genetic fingerprints, pedigrees, and their associations to diseases and other complex traits; application to clinical diagnosis and forensic analysis

o   Information and Intelligent Systems: with emphasis on developing web-scale, multi-core, and on-line machine learning systems for social media, computer vision, and HCI applications. Of particular interest are:

1)     Multi-view latent space models, topics models, sparse coding methods for image/text/relational information retrieval

2)     Evolving structure, stable metrics, and prediction for large-scale dynamic social networks; goal-driven network design/modification/optimization

3)     eb-scale image understanding, search, annotation, and retrieval; photo storyline; analysis of video and multimedia

4)     User modeling and personalization, computational advertising, and temporal analysis based on image, text, and activities