DERI Seminar with Dr Kostas Papafitsoros
When: Thursday, March 16, 2023, 11:00 AM - 12:00 PM
Where: Zoom
Speaker: Dr Kostas Papafitsoros, Lecturer in Mathematical Data Science at QM
ZOOM link: https://qmul-ac-uk.zoom.us/j/81148100921
Title: Deep learning and mathematical modelling for sea turtle research and conservation: Status, challenges and opportunities
Abstract: I will initially give a general overview of how deep learning and mathematical modelling combined with large scale imaging data have been so far beneficial to sea turtle research and conservation. I will focus on tasks like individual animal re-identification, animal detection, behavioural classification and measuring tourist pressure using social media data. I will then introduce SeaTurtleID, a novel long-span publicly available dataset of sea turtle photographs captured in the wild, which is suitable for benchmarking re-identification and evaluating several other computer vision tasks. I will finish with a discussion about further potential applications of deep learning to sea turtle research that can be facilitated by this dataset, a discussion which can also serve as a starting point for establishing new collaborations.
Profile: I am a Lecturer in Mathematical Data Science at the School of Mathematical Sciences, Queen Mary University of London. Prior to that (2017-2022), I was a research scientist at the Weierstrass Institute for Applied Analysis and Stochastics, in Berlin, Germany, working in the group Nonsmooth Variational Problems and Operator Equations. During the period September 2015-August 2017, I was an Alexander von Humboldt Postdoctoral Fellow, working initially at the Mathematical Institute of Humboldt University Berlin and later at the Weierstrass Institute. I completed my PhD in 2014 at the University of Cambridge, where I was also a member of the Cambridge Image Analysis group. I stayed in Cambridge six more months after my PhD, as an EPSRC Doctoral Prize fellow at the Department of Applied Mathematics and Theoretical Physics.
My main research area is mathematical imaging, in the interface of several areas of applied mathematics, like inverse problems, variational methods, calculus of variations, functional analysis, optimisation, optimal control, numerical analysis and deep learning. I have been particularly involved in the design, analysis and application of nonsmooth energy functionals incorporated in variational regularisation methods tailored for image processing. These functionals are at one hand discontinuity-preserving often stemming from their nonsmoothness, that is, they have the ability to preserve sharp edges in the images and on the other hand, they are adaptive to the specific structure of the given data. I am interested in combining these "classical" methods with modern data-driven and deep learning approaches.
I have a second research direction which is the result of my long-term involvement in environmental conservation with special focus on marine turtles, see the corresponding section of my personal website. I consider several applications of animal photo-identification (identification of individual animals via their unique morphological characteristics e.g. patterns/features), and in the same time consider various ways to enhance and automatise this process via modern imaging techniques. I am also looking at mathematical/statistical aspects of citizen science projects related to wildlife conservation.