Robust generalisation
Understand when robustness transfers beyond the training sample—and why accuracy and robustness sometimes compete.
Research programme · 01
Learning Theory: Trustworthy AI Foundations
What makes learning reliable under adversarial data, supervision, and evaluation?
Core algorithms and theory for trustworthy machine learning, from adversarial training and robust generalisation to representation alignment.
6 selected research outputs5 directly related talks
Research focus
Three questions shape this programme.
This is the working agenda: a precise statement of the problems the lab is building toward, not a claim that every component is already solved.
Reliable representations
Study how contrastive and foundation-model representations become aligned, stable, and diagnostically useful.
Principled intervention
Turn theoretical failure analysis into training objectives, sample-selection rules, and measurable guarantees.
Selected publications & outcomes
Evidence behind the direction.
Peer-reviewed papers and clearly labelled arXiv preprints are shown with their source records. See the complete Google Scholar profile ↗ for the full publication history.
Benign Overfitting in Adversarial Training for Vision Transformers
Studies benign overfitting and generalisation in adversarially trained vision transformers.
Primary source arXivDissecting Representation Misalignment in Contrastive Learning via Influence Function
Uses influence-function analysis to examine representation misalignment in contrastive learning.
Primary sourceOne Stone, Two Birds: Enhancing Adversarial Defense Through the Lens of Distributional Discrepancy
Approaches adversarial defence through the distributional discrepancy between clean and adversarial data.
Primary source CodeImproving Accuracy-Robustness Trade-off via Pixel Reweighted Adversarial Training
Studies the accuracy–robustness trade-off through pixel-reweighted adversarial training.
Primary source CodeEfficient Adversarial Contrastive Learning via Robustness-Aware Coreset Selection
Makes adversarial contrastive learning more efficient with robustness-aware coreset selection.
Primary source CodeGeometry-aware Instance-reweighted Adversarial Training
Uses instance geometry to reweight adversarial training; selected for an ICLR Oral.
Primary source CodeRelated talks
The programme, explained in public.
Selected talks that develop or communicate this research direction.
Attack and Defense Mechanisms in Trustworthy Machine Learning: Challenges and Future Directions
Research Awakening · AI Frontier Lecture — PaperWeekly × STARC · Online (China)
Robust Machine Learning in Adversarial Environments
8th Guanghua Young Scholars Forum, Institute of Big Data, Fudan University · Online · Shanghai, China
Adversarial Learning: Foundations and Applications
BKAI, Hanoi University of Science and Technology · Hanoi, Vietnam
Adversarial Robustness: From Basic Science to Applications
EPFL-CIS & RIKEN-AIP Joint Online Workshop on Machine Learning · Online
Video ↗Geometry-aware Instance-reweighted Adversarial Training
ICLR 2021 · Online
The full agenda
Explore all four research layers.
Students · collaborators · partners
Help move this programme from a hard question to reliable practice.
RISE-AI welcomes people who value rigorous foundations, reproducible systems, and generous teamwork.
See openings and how to apply →