Welcome back to Beyond the Patterns. Today I have the great pleasure to announce Rekha
Vendra Selvan, who is currently an assistant professor at the University of Copenhagen.
He has responsibilities at the machine learning section, the Department of Computer Science,
the Keen Lab, the Department of Neuroscience and the Data Science Laboratory. He received
his PhD in Medical Image Analysis also at the University of Copenhagen and his master's
degree in Communication Engineering in 2015 at the Chalmers University in Sweden. His
bachelor degree he obtained in Electronics and Communication Engineering in 2009 at the
BMS Institute of Technology in India. Rekha Vendra was born in Bangalore, India. His current
research interests are broadly pertaining to medical image analysis using quantum tensor
networks, resource efficient machine learning, Bayesian machine learning, graph neural networks,
approximate inference and multi-object tracking theory. So today I have the great pleasure
to announce his presentation entitled Quantum Tensor Networks for Medical Image Analysis.
So Rekha, it's a great pleasure to have you here and I'm very much looking forward to
your presentation. So the stage is yours.
Andreas for the very kind introduction and also the opportunity to virtually come and
talk to this crowd of your students, I assume. So yeah, so you can call me Raghav in short
and yeah, I think the introduction was basically covers everything that I wanted to say. But
if you are on Twitter, so you can follow me at that particular handle, that's all my title
slide is for. So the talk today is going to be about quantum tensor networks for medical
image analysis. So this is kind of converging two of my core interests with a recent interest
that I have had with quantum tensor networks. And I hope by the end of this talk, you will
also be excited as much as I am about this particular tool. So just before that, a bit
more about my current research as to what I'm doing. So my core research area is medical
image analysis. So that's where I have my PhD from and since that was in 2015. So since
then I've still continued to do more of medical image analysis using broadly machine learning
and then like last year we had some bit of work using the Danish cohort of the first
wave of COVID where we also try to predict the risk in terms of hospitalization. And
so that was like a multi institutional effort. But it was pretty exciting to see sort of
my research in some sense being used in like a real life application that we are all still
going through. So other than medical image analysis, I also because of my affiliation
with the data science lab at the University of Copenhagen, I get to collaborate with different
departments and then do really exciting research, I think applied but still very exciting. So
for instance, we were trying to characterize the behavior of narvials in the Arctic using
machine learning. And then we are trying to kind of look at the diversity of insects in
Sweden and then obtain properties of so obtain molecules, nanoparticles of a particular desired
structure based on certain properties. And my work at the Kean lab, which is my other
half where I'm affiliated with the neuroscience department, we study the brain. So where we're
trying to characterize the behavior of mammals and then trying to look at the neural neural
circuits there. So it's kind of like a very diverse but very gratifying sort of set of
applications that I get to work with. So further, there are like few other interests, which
also certainly helped my research. So if you are ever in Copenhagen, and then if you would
like to take a tour of the trails around, so I'm into trail running, but if you're not
into running, so just drop me a message and then we can go for a hike also where we can
discuss some research. So this, with this background, here's sort of the overview of
today's talk. So I think there's going to be a bit of, you know, background that has
to be provided with these tensor networks. So I think I'm going to use the first 10 to
15 minutes trying to motivate and then provide basic sort of notations and then some of the
framework that will be used in sort of the two sets of applications that we're going
to look at. So it's tensor networks for first, we look at classification and then how we
can extend them for segmentation. So that's sort of the overview. And say, like if there
Presenters
Zugänglich über
Offener Zugang
Dauer
01:11:46 Min
Aufnahmedatum
2021-05-26
Hochgeladen am
2021-05-27 00:49:40
Sprache
en-US
It’s a great pleasure to welcome Raghav Selvan from the University of Copenhagen at our lab!
Abstract: Quantum Tensor Networks (QTNs) provide efficient approximations of operations involving high dimensional tensors and have been extensively used in modeling quantum many-body systems and also compressing large neural networks. More recently, supervised learning has been attempted with tensor networks, and has primarily focused on classification of 1D signals and small images. In this talk, we will look at two formulations of QTN-based models for 2D & 3D medical image classification and 2D medical image segmentation. Both the classification and segmentation models use the matrix product state (MPS) tensor network under the hood, which efficiently learns linear decision rules in high dimensional spaces. These QTN models are fully linear, end-to-end trainable using backpropagation, and have a lower GPU memory footprint than convolutional neural networks (CNN). We show competitive performance compared to relevant CNN baselines on multiple datasets for classification and segmentation tasks while presenting interesting connections to other existing supervised learning methods.
Bio: Raghavendra Selvan (Raghav) is currently an Assistant Professor at the University of Copenhagen, with joint responsibilities at the Machine Learning Section (Dept. of Computer Science), Kiehn Lab (Department of Neuroscience), and the Data Science Laboratory. He received his Ph.D. in Medical Image Analysis (University of Copenhagen, 2018), his MSc degree in Communication Engineering in 2015 (Chalmers University, Sweden), and his Bachelor's 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 to Medical Image Analysis using Quantum Tensor Networks, Resource-efficient ML, Bayesian Machine Learning, Graph-neural networks, Approximate Inference and multi-object tracking theory.
References
Raghav's Website https://di.ku.dk/english/staff/?pure=en/persons/532407
Raghav's Github Page https://raghavian.github.io
Slides: https://raghavian.github.io/talks/files/FAU_20210526.pdf
Tensor Networks for Medical Image Classification (2020) http://proceedings.mlr.press/v121/selvan20a.html
Locally orderless tensor networks for classifying two- and three-dimensional medical images (2021) https://www.melba-journal.org/article/21663-locally-orderless-tensor-networks-for-classifying-two-and-three-dimensional-medical-images?auth_token=HgMd7jGPhvS8EqDEmj30
Multi-layered tensor networks for image classification (2020) https://arxiv.org/abs/2011.06982
Segmenting two-dimensional structures with strided tensor networks (2021) https://arxiv.org/abs/2102.06900
Classification model: https://github.com/raghavian/loTeNet_pytorch/
Segmentation model: https://github.com/raghavian/strided-tenet
This video is released under CC BY 4.0. Please feel free to share and reuse.
For reminders to watch the new video follow on Twitter or LinkedIn. Also, join our network for information about talks, videos, and job offers in our Facebook and LinkedIn Groups.
Music Reference:
Damiano Baldoni - Thinking of You (Intro)
Damiano Baldoni - Poenia (Outro)