Skip to main content
Events
  1. Queen Mary University of London
  2. Events
  3. Past events
  4. 2022
  5. Items

DERI Seminar with Dr Ozan Oktay, Principal Researcher at Microsoft Research Cambridge

When: Thursday, April 7, 2022, 11:00 AM - 12:00 PM
Where: remote zoom

Speaker: Dr Ozan Oktay, Principal Researcher at Microsoft Research Cambridge

Title: Learning with imperfect datasets under resource constraints: How to curate the right labels and avoid potential biases in model selection

https://qmul-ac-uk.zoom.us/j/81148100921

Abstract:  Imperfections in data annotation, known as label noise, are detrimental to the training of machine learning models and have an often-overlooked confounding effect on the assessment of model performance. Nevertheless, employing experts to remove label noise by fully re-annotating large datasets is infeasible in resource-constrained settings, such as healthcare. In this talk, I will present our recent work on "active label cleaning" (see manuscript), a data-driven approach to prioritising samples for re-annotation. We propose to rank instances according to estimated label correctness and labelling difficulty of each sample, and introduce a simulation framework to evaluate relabelling efficacy. Our experiments on natural images and on a new medical imaging benchmark show that cleaning noisy labels mitigates their negative impact on model training, evaluation, and selection. Crucially, the proposed active label cleaning enables correcting labels up to 4 times more effectively than typical random selection in realistic conditions, making better use of experts' valuable time for improving dataset quality.

Short Bio:Ozan Oktay is a Principal Researcher at Microsoft Research Cambridge, where he is leading the research efforts in Medical Imaging team (formerly known as Project InnerEye). His research work focusses on developing trustworthy and reliable machine learning (ML) for digital health applications, including robust representation learning, data-efficient learning, and visual-language processing. Prior to joining Microsoft Research, Dr Oktay involved in several early-to-late-stage med-tech companies to advance their research over the years, including HeartFlow Inc  (CA, USA), Siemens Corporate Research (NJ, USA), and ThinkSono Ltd (UK). He is currently affiliated with the Computing Department at Imperial College London as Honorary Research Fellow, where he held a Research Associate role during his PhD study with Prof. Daniel Rueckert and Dr Ben Glocker.

https://www.microsoft.com/en-us/research/people/ozoktay/

 

 

 

 

 

Back to top