Julia A. Meister

News:

• Aug 2022 -
  • Proceedings paper "Cough-based COVID-19 detection with audio quality clustering and confidence measure based learning". Presented at the COPA2022 conferece, where it won the "Best Student Paper" award.
  • Proceedings paper "Audio feature ranking for sound-based COVID-19 patient detection". Presented at the EPIA2022 conference.
• June 2022 - Winner of the "3 Minute Thesis" (3MT) competition at University of Brighton, progressing to the UK national competition.
• Apr 2022 - Research Officer, developed a novel uncertainty quantification methods for Deep Transfer Learning.
• Feb 2022 - Research Officer, evaluated the carbon footprint of remanufactured catheters. Collaborated closely with the NHS and two US-based medical device remanufacturers: AMDR and Innovative Health.
• Oct 2021 - Presented in-progress work on Machine Learning COVID-19 cough detection at the IRSH2021 conference.
• July 2021 - Supervised and supported a Brighton undergraduate student's research activities as part of the Santander-funded "Global Challenges" grant.
• Apr 2021 - 1 of 7 winners of the Santander-funded "Global Challenges" grant, in collaboration with Dr Khuong Nguyen, Prof Zhiyuan Luo, and Cardisio (Germany), an industry practitioner specialised in Machine Learning for healthcare.
• Jan 2021 - Joined University of Brighton as a PhD student.
• Dec 2020 - Awarded MSc in Data Science and Analytics, thesis titled "Conformal Predictors for detecting harmful respiratory events".

After completing BSc and MSc degrees at Royal Holloway, University of London, I joined Brighton University's School of Computing, Engineering & Maths. Under the supervision of Dr Khuong Nguyen, Dr Marcus Winter, and Prof Alison Bruce, I am pursuing a PhD focused on developing uncertainty quantification and Confidence Machine Learning methods for Digital Health.

Given a Machine Learning prediction, how confident are you that it is correct?

Often, we assume that models will be about as successful on future samples as they are on their training data. Without uncertainty quantification, ML's application in the safety-critical healthcare setting can be unreliable.

In contrast, Confidence ML associates individual predictions with confidence levels to provide guaranteed error probabilities per sample. My research centres on developing and improving ML techniques to provide confident and efficient predictions for Digital Health applications.

Recent projects include detecting COVID-19 infections from respiratory audio, and identifying the risk of coronary heart disease from EKG sensor data.

Keywords: Digital Health • Confidence Machine Learning • Data Science and Analytics