Poisoning & backdoors
Reveal how manipulated samples, labels, and physical triggers can redirect learning or downstream behaviour.
Research programme · 02
AI Governance: Data Safety & Content Integrity
How do we keep data-driven systems reliable when data can be poisoned, biased, noisy, or private?
Security and governance at the data layer: poisoning and backdoors, algorithmic bias, privacy leakage, and defensible data use.
6 selected research outputs2 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.
Bias & fair generation
Measure and mitigate systematic disparities in generative and predictive systems.
Privacy-preserving adaptation
Reduce information leakage while retaining the practical value of fine-tuning and collaborative learning.
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.
MP-Nav: Enhancing Data Poisoning Attacks against Multimodal Learning
Investigates data-poisoning attacks against multimodal learning systems.
Primary sourceFair Text-to-Image Diffusion via Fair Mapping
Introduces fair mapping as a route to more equitable text-to-image diffusion.
Primary source arXivPrivacy-Preserving Low-Rank Adaptation Against Membership Inference Attacks for Latent Diffusion Models
Protects low-rank adaptation of latent diffusion models against membership-inference attacks.
Primary source CodeLearning a Robust Foundation Model Against Clean-Label Data Poisoning Attacks at Downstream Tasks
Studies foundation-model robustness to clean-label poisoning in downstream tasks.
Primary sourcePrivacy-Preserving Heterogeneous Federated Learning
Addresses privacy in heterogeneous federated learning.
Primary sourceBilateral Dependency Optimization: Defending Against Model-Inversion Attacks
Defends against model-inversion attacks through bilateral dependency optimisation.
Primary sourceRelated talks
The programme, explained in public.
Selected talks that develop or communicate this research direction.
Privacy-Preserving Adaptation against Membership Inference Attacks for Latent Diffusion Models
International Workshop on Weakly Supervised Learning 2026 · Seoul, South Korea
Evaluating and Enhancing the Reliability of Semi-Supervised Learning
Joint Workshop on Science & Technology Hub Collaborative Research Program · Kobe, Japan
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 →