Raghavendra Selvan (Raghav) is currently a Postdoctoral researcher at the Machine Learning Section at the Department of Computer Science and the Data Science Lab of University of Copenhagen. He received his PhD in Medical Image Analysis (University of Copenhagen, 2018), his MSc degree in Communication Engineering in 2015 (Chalmers University, Sweden) and his Bachelor degree in Electronics and Communication Engineering degree in 2009 (BMS Institute of Technology, India). Raghavendra Selvan was born in Bangalore, India.
His current research interests are broadly pertaining Bayesian Machine Learning, with a focus on Medical image analysis, Graph-based learning, Tensor networks, Approximate Bayesian inference and Bayesian tracking theory.
- 2020-08: Featured in Computer Vision News with an interview about our MIDL paper
- 2020-07: Runner-up for the Best Paper Award at MIDL-2020
- 2019-01: Started as Post-doc at Dept. of Computer Science, University of Copenhagen
- 2018-11: Successfully defended my PhD thesis
(Switching to first person :)
I am actively involved in a wide range of research projects as part of my roles at the Machine Learning section and Data Science Lab of University of Copenhagen. Brief descriptions of ongoing and completed projects are listed below.
COVID-19 Risk Modelling: As part of a collaboration with the hospitals in Zealand region of Denmark, we are modelling COVID-19 risk from clinical and image data. I am contributing to the team focusing on using chest X-rays at admission to model risk of COVID-19.
Uncertainty Quantification: Medical image segmentation with well calibrated and meaningful uncertainty estimation can be more useful in clinical settings. For example, we present a probabilistic segmentation model that uses normalizing flows to improve the diversity of segmented samples.
Low Resource Machine Learning: Reducing the reliance on high quality labeled data and expensive hardware resources can globally democratise ML methods. Addressing one of these factors, we recently introduced tensor networks for medical image classification tasks that require less than 10% of GPU resources when compared to traditional deep learning models. We have also put out a tool – Carbontracker – to predict and track carbon footprint of developing deep learning models, in an effort to encourage research into energy-efficient models.
Locomotion analysis: In this collabortion with Kiehn Lab, my role has been to analyze videos of mice running on treadmills in order to extract speed, cadence and coordination statistics. The results from analyzing some 1000 videos have accentuated the role of specific interneurons in the spinal cord that influence locomotion.
Generative Modelling of nanoparticles: Obtaining structures from scattering data for nanomolecules turns out to be an expensive and difficult task. This ongoing collaboration with Prof. Kirsten Jensen’s group we are explorting generating valid nanomolecular structures from x-ray scattering data using deep generative models with promising early results.
Behavioural analysis of narwhals: Echolocation used by narwhals can reveal their behaviour. In this collaboration with Prof. Susanne Ditlevsen’s group, we are using machine learning methods to decode the echolation data and understand the animal behaviour.
- Segmentation of roots in soil (2019): Collaboration with Plant & Environ. Science Dept.
- Extraction of airways from volumetric data (2018): PhD Thesis
- Bayesian tracking of multiple point targets (2015): MSc Thesis
- India’s first pico-satellite (2010): Collaboration with Indian Space Research Organisation