Imperfect Supervision
Noisy labels, open-set noise, coarse labels, and reliable representation learning from imperfect visual data.
Queen's University Belfast
Robust machine learning for imperfect information.
We study machine learning and computer vision systems that remain reliable when supervision, data, evaluators, or deployment conditions are noisy, incomplete, or adversarial.
Research Focus
QUB ML Lab works on robust machine learning, AI safety, computer vision, and foundation models when data, labels, evaluators, or deployment conditions are imperfect.
AI Safety and Evaluation
Certification, adversarial robustness, and statistically reliable evaluation for models and LLM-based systems.
Foundation Models
Robust visual and multimodal learning with CLIP-style representations, healthcare AI, and domain-shifted data.
Latest News
- Oct 2026 AGraphHF, led by PhD student Tadiyos Hailemichael Mamo, was published in Artificial Intelligence in Medicine.
- Oct 2026 The lab received Crucible pump-priming funding for TRUST-H2, together with 10,000 GPU hours each on the UKRI AI Research Resource (Zenith) and Isambard-AI.
- Oct 2026 Dr. Feng joined IEEE Computational Intelligence Letters as an Associate Editor and became a Fellow of the Higher Education Academy.
- Sep 2026 ViewSAM, on weakly supervised cross-view referring multi-object tracking, was accepted to NeurIPS 2026.
- Sep 2026 Zhengli Shi and Zihan Xu joined the lab as PhD students, working on multimodal safety and on foundation models for multimodal data.
- Aug 2026 Dr. Feng is co-organising the NeurIPS 2026 Workshop on Embodied and Safe-Assured Robotic Systems.
- Jul 2026 Dr. Feng concluded a research visit to the KUIS AI Center at Koç University, hosted by Prof. Çiğdem Gündüz Demir, running from May to July 2026.
- Jun 2026 “Deconstructing the Failure of Ideal Noise Correction” was presented at CVPR 2026.
- Mar 2026 Dr. Feng received the Journal of Cybersecurity and Privacy Travel Award.
- Feb 2026 Two first-author papers were accepted to ICLR 2026 and CVPR 2026.
- Oct 2025 Two papers were accepted to ACM MM 2025, including work on open-set label noise.
- Feb 2025 PROSAC was accepted by AAAI 2025 as an oral presentation.