Mr Demetris Lappas
Contact
- Email: k1838447@kingston.ac.uk
- Location: Penrhyn Road
- Faculty: Faculty of Engineering, Computing and the Environment
- School: School of Computer Science and Mathematics
- Department: Department of Computer Science
Research project: Anomaly Detection in Computer Vision using Deep Learning Models
Abstract
Anomaly detection involves identifying rare patterns in data that deviate from expected behavior. Since anomalies are, by definition, infrequent, automated detection mechanisms are invaluable, offering the potential to reduce human oversight, minimize errors, and lower operational costs.
Traditional machine learning methods often fall short in computer vision, where detecting context-dependent anomalies in images and videos requires more advanced techniques. Standard reconstruction-based approaches, while commonly used, are fundamentally similar to dimensionality reduction algorithms, which can inadvertently learn to reconstruct anomalies as well as normal patterns. My research addresses this limitation by introducing dynamic anomaly weighting, pseudo anomalies, and novel architectures such as masked convolutions within skip connections. These innovations promote the explicit transformation of anomalies back into normal patterns, rather than merely reconstructing them.
- Research degree: PhD
- Title of project: Anomaly Detection in Computer Vision using Deep Learning Models
- Research supervisor: Professor Dimitrios Makris
- Other research supervisor: Professor Vasilis Argyriou
Biography
I am a PhD student specializing in anomaly detection in computer vision using deep learning models, and I expect to complete my doctorate in November 2024. My research focuses on developing novel architectures that infer anomalous spatio subspaces by leveraging surrounding normal spatio-temporal subspaces in images and videos. I have also designed an innovative loss function that trains a reconstruction model with a dynamic anomaly weight, transforming anomalous features into normal ones to amplify reconstruction loss.
In addition to my academic pursuits, I am a full-time Senior Director of AI and Data Science in the insurance sector, where I lead a team developing machine learning and deep learning solutions. Prior to this, I led a research team at Aspen Technologies, exploring how deep learning models can mimic first-principle equations while satisfying physical constraints. My professional experience also includes roles at Barclays Bank and Ernst & Young in departments such as Cards and Payments, Financial Crime, and Entity Resolution & Data Enrichment.
Areas of research interest
- Artificial Intelligence
- Computer Vision
- Anomaly Detection
- Machine Learning
- Deep Learning
- Geometric Deep Learning
Qualifications
- BSc in Mathematics, London Metropolitan University
- Postgraduate in Mathematics, University of Glasgow
Publications
- Dynamic Distinction Learning: Adaptive Pseudo Anomalies for Video Anomaly Detection – In: IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR) 2024, Seattle, U.S. (https://openaccess.thecvf.com/content/CVPR2024W/VAND/html/Lappas_Dynamic_Distinction_Learning_Adaptive_Pseudo_Anomalies_for_Video_Anomaly_Detection_CVPRW_2024_paper.html)
- Masked Convolutions in Skip Connections for Enhanced Video Anomaly Detection – Preprint on SSRN, 2023.(https://papers.ssrn.com/sol3/papers.cfm?abstract_id=4768663)
- Enforced Isolation Deep Networks for Anomaly Detection in Images – In: IEEE/CVF International Conference on Computer Vision Workshops (ICCVW) 2021, Montreal, Canada.(https://ieeexplore.ieee.org/abstract/document/9569000)
- Fourier Transformation Autoencoders for Anomaly Detection – In: IEEE International Conference on Acoustics, Speech, and Signal Processing (ICASSP) 2021, Toronto, Ontario, Canada. (https://ieeexplore.ieee.org/abstract/document/9415010)
Jump to: Conference or Workshop Item
Number of items: 2.
Conference or Workshop Item
Lappas, Demetris, Argyriou, Vasileios and Makris, Dimitrios (2024) Dynamic distinction learning : adaptive pseudo anomalies for video anomaly detection. In: The IEEE/CVF Conference on Computer Vision and Pattern Recognition 2024; 17-21 June 2024, Seattle, U.S..
Lappas, Demetris, Argyriou, Vasileios and Makris, Dimitrios (2021) Fourier transformation autoencoders for anomaly detection. In: 2021 IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP); 06 - 11 Jun 2021, Toronto, Ontario, Canada (Held online).