Imperial College London

Dr Arinbjörn Kolbeinsson

Faculty of MedicineSchool of Public Health

Honorary Research Associate



Royal School of MinesSouth Kensington Campus





Publication Type

7 results found

Lightley J, Gorlitz F, Kumar S, Kalita R, Kolbeinsson A, Garcia E, Alexandrov Y, Bousgouni V, Wysoczanski R, Barnes P, Donnelly L, Bakal C, Dunsby C, Neil MAA, Flaxman S, French PMWet al., 2021, Robust deep learning optical autofocus system applied to automated multiwell plate single molecule localization microscopy, JOURNAL OF MICROSCOPY, ISSN: 0022-2720

Journal article

Kolbeinsson A, Kossaifi J, Panagakis I, Bulat A, Anandkumar A, Tzoulaki I, Matthews Pet al., 2021, Tensor dropout for robust learning, IEEE Journal of Selected Topics in Signal Processing, Vol: 15, Pages: 630-640, ISSN: 1932-4553

CNNs achieve high levels of performance by leveraging deep, over-parametrized neural architectures, trained on large datasets. However, they exhibit limited generalization abilities outside their training domain and lack robustness to corruptions such as noise and adversarial attacks. To improve robustness and obtain more computationally and memory efficient models, better inductive biases are needed. To provide such inductive biases, tensor layers have been successfully proposed to leverage multi-linear structure through higher-order computations. In this paper, we propose tensor dropout, a randomization technique that can be applied to tensor factorizations, such as those parametrizing tensor layers. In particular, we study tensor regression layers, parametrized by low-rank weight tensors and augmented with our proposed tensor dropout. We empirically show that our approach improves generalization for image classification on ImageNet and CIFAR-100. We also establish state-of-the-art accuracy for phenotypic trait prediction on the largest available dataset of brain MRI (U.K. Biobank), where multi-linear structure is paramount. In all cases, we demonstrate superior performance and significantly improved robustness, both to noisy inputs and to adversarial attacks. We establish the theoretical validity of our approach and the regularizing effect of tensor dropout by demonstrating the link between randomized tensor regression with tensor dropout and deterministic regularized tensor regression.

Journal article

Kolbeinsson A, Filippi S, Panagakis I, Matthews P, Elliott P, Dehghan A, Tzoulaki Iet al., 2020, Accelerated MRI-predicted brain ageing and its associations with cardiometabolic and brain disorders, Scientific Reports, Vol: 10, ISSN: 2045-2322

Brain structure in later life reflects both influences of intrinsic aging and those of lifestyle, environment and disease. We developed a deep neural network model trained on brain MRI scans of healthy people to predict “healthy” brain age. Brain regions most informative for the prediction included the cerebellum, hippocampus, amygdala and insular cortex. We then applied this model to data from an independent group of people not stratified for health. A phenome-wide association analysis of over 1,410 traits in the UK Biobank with differences between the predicted and chronological ages for the second group identified significant associations with over 40 traits including diseases (e.g., type I and type II diabetes), disease risk factors (e.g., increased diastolic blood pressure and body mass index), and poorer cognitive function. These observations highlight relationships between brain and systemic health and have implications for understanding contributions of the latter to late life dementia risk.

Journal article

Lengyel D, Petangoda J, Falk I, Highnam K, Lazarou M, Kolbeinsson A, Deisenroth MP, Jennings NRet al., 2020, GENNI: Visualising the geometry of equivalences for neural network identifiability, Publisher: arXiv

We propose an efficient algorithm to visualise symmetries in neural networks.Typically, models are defined with respect to a parameter space, wherenon-equal parameters can produce the same input-output map. Our proposedmethod, GENNI, allows us to efficiently identify parameters that arefunctionally equivalent and then visualise the subspace of the resultingequivalence class. By doing so, we are now able to better explore questionssurrounding identifiability, with applications to optimisation andgeneralizability, for commonly used or newly developed neural networkarchitectures.

Working paper

Kossaifi J, Lipton ZC, Kolbeinsson A, Khanna A, Furlanello T, Anandkumar Aet al., 2020, Tensor Regression Networks, JOURNAL OF MACHINE LEARNING RESEARCH, Vol: 21, ISSN: 1532-4435

Journal article

Bintsi KM, Baltatzis V, Kolbeinsson A, Hammers A, Rueckert Det al., 2020, Patch-Based Brain Age Estimation from MR Images, Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics), Vol: 12449 LNCS, Pages: 98-107, ISSN: 0302-9743

Brain age estimation from Magnetic Resonance Images (MRI) derives the difference between a subject’s biological brain age and their chronological age. This is a potential biomarker for neurodegeneration, e.g. as part of Alzheimer’s disease. Early detection of neurodegeneration manifesting as a higher brain age can potentially facilitate better medical care and planning for affected individuals. Many studies have been proposed for the prediction of chronological age from brain MRI using machine learning and specifically deep learning techniques. Contrary to most studies, which use the whole brain volume, in this study, we develop a new deep learning approach that uses 3D patches of the brain as well as convolutional neural networks (CNNs) to develop a localised brain age estimator. In this way, we can obtain a visualization of the regions that play the most important role for estimating brain age, leading to more anatomically driven and interpretable results, and thus confirming relevant literature which suggests that the ventricles and the hippocampus are the areas that are most informative. In addition, we leverage this knowledge in order to improve the overall performance on the task of age estimation by combining the results of different patches using an ensemble method, such as averaging or linear regression. The network is trained on the UK Biobank dataset and the method achieves state-of-the-art results with a Mean Absolute Error of 2.46 years for purely regional estimates, and 2.13 years for an ensemble of patches before bias correction, while 1.96 years after bias correction.

Journal article

Richemond PH, Kolbeinsson A, Guo Y, Biologically inspired architectures for sample-efficient deep reinforcement learning

Deep reinforcement learning requires a heavy price in terms of sampleefficiency and overparameterization in the neural networks used for functionapproximation. In this work, we use tensor factorization in order to learn morecompact representation for reinforcement learning policies. We show empiricallythat in the low-data regime, it is possible to learn online policies with 2 to10 times less total coefficients, with little to no loss of performance. Wealso leverage progress in second order optimization, and use the theory ofwavelet scattering to further reduce the number of learned coefficients, byforegoing learning the topmost convolutional layer filters altogether. Weevaluate our results on the Atari suite against recent baseline algorithms thatrepresent the state-of-the-art in data efficiency, and get comparable resultswith an order of magnitude gain in weight parsimony.

Journal article

This data is extracted from the Web of Science and reproduced under a licence from Thomson Reuters. You may not copy or re-distribute this data in whole or in part without the written consent of the Science business of Thomson Reuters.

Request URL: Request URI: /respub/WEB-INF/jsp/search-html.jsp Query String: respub-action=search.html&id=00690138&limit=30&person=true