Nikolaos (Nikos) Kargas

Ph.D. Student - Research Assistant
Department of Electrical and Computer Engineering
University of Minnesota
Email: karga005 at

About me

I am a Ph.D. student in the Department of Electrical and Computer Engineering, at the University of Minnesota advised by Professor Nikolaos D. Sidiropoulos. My research interests are in the areas of Machine learning, Statistics and Optimization.

A major focus of my work is on tensor methods for machine learning. I am interested in developing scalable, efficient and interpretable algorithms for high-dimensional distribution and general nonlinear function learning. Recent research topics include non-parametric density estimation, learning probability models from limited and partially observed data, data completion/regression and spatio-temporal data analysis.

I received the Diploma and MSc degrees from the School of Electronic & Computer Engineering at the Technical University of Crete. As an undergraduate student, I was a member of the RoboCup team Kouretes working on the team's localization module. As an MSc student, I was a member of the Telecom Lab working on backscatter networks for large-scale environmental sensing.



asilomar2020 Supervised Learning via Ensemble Tensor Completion
N. Kargas and N. D. Sidiropoulos
Asilomar Conference on Signals, Systems and Computers (ASILOMAR), 2020

Our recent work has formulated the problem of learning a general nonlinear multivariate function of discrete inputs, as a tensor completion problem with smooth latent factors. In this work, we utilize two ensemble learning techniques to enhance its prediction accuracy.
aaai2020 Nonlinear System Identification via Tensor Completion
N. Kargas and N. D. Sidiropoulos
AAAI Conference on Artificial Intelligence (AAAI), 2020 (Spotlight)
We show that identifying a general nonlinear function from input-output examples can be formulated as a tensor completion problem and under certain conditions provably correct nonlinear system identification is possible. We extend our method to the multi-output setting and the case of partially observed data.
tci2019 Low-Rank Tensor Models for Improved Multi-Dimensional MRI: Application to Dynamic Cardiac T1 Mapping
B. Yaman, S. Weingartner, N. Kargas, N. D. Sidiropoulos and M. Akcakaya
IEEE Transactions on Computational Imaging, 2019

We explore different tensor decomposition methods in order to enable high-resolution cardiac phase-resolved myocardial T1 mapping.
dsw2019 Statistical Learning Using Hierarchical Modeling of Probability Tensors
M. Amiridi, N. Kargas and N. D. Sidiropoulos
IEEE Data Science Workshop (DSW), 2019 (Best student paper award)
We address the complexity of accurately estimating high-dimensional joint distributions by proposing a novel hierarchical learning algorithm for probability mass function (PMF) estimation through parallel local views of a probability tensor.
neurips2019 Crowdsourcing via Pairwise Co-occurrences: Identifiability and Algorithms
S. Ibrahim, X. Fu, N. Kargas and K. Huang
Advances in Neural Information Processing Systems (NeurIPS), 2019
We propose a framework using pairwise co-occurrences of the annotator responses and show that our approach can identify the Dawid-Skene model under realistic conditions. We develop two algorithms to solve the model identification problem.
aistats2019 Learning Mixtures of Smooth Product Distributions: Identifiability and Algorithm
N. Kargas and N. D. Sidiropoulos
International Conference on Artificial Intelligence and Statistics (AISTATS), 2019
We study the problem of learning a mixture model of non-parametric product distributions and propose a two-stage approach which recovers the component distributions of the mixture.
tsp2018 Tensors, Learning, and `Kolmogorov Extension' for Finite-alphabet Random Vectors
N. Kargas, N. D. Sidiropoulos and X. Fu
IEEE Transactions on Signal Processing, 2018
We prove that high-dimensional PMF recovery from low-dimensional marginalized PMFs can be guaranteed under certain low-rank conditions. We derive identifiability results and an algorithm to carry out the recovery task.
camsap2017 Locally Low-Rank tensor regularization for high-resolution quantitative dynamic MRI
B. Yaman, S. Weingartner, N. Kargas, N. D. Sidiropoulos and M. Akcakaya
IEEE 7th International Workshop on Computational Advances in Multi-Sensor Adaptive Processing (CAMSAP), 2017

We propose a locally low-rank tensor regularization approach for high-resolution quantitative dynamic MRI.
spars2017 Low-Rank Tensor Regularization for Improved Dynamic Quantitative Magnetic Resonance Imaging
N. Kargas, S. Weingartner, N. D. Sidiropoulos and M. Akcakaya
Signal Processing with Adaptive Sparse Structured Representations Workshop (SPARS), 2017

We propose a method based on low-rank tensor regularization for improved undersampled dynamic quantitative MRI reconstruction.
ita2017 Completing a Joint PMF from Projections: a Low-rank Coupled Tensor Factorization Approach
N. Kargas and N. D. Sidiropoulos
Information Theory and Applications Workshop (ITA), 2017
We show that it is possible to recover higher-dimensional joint PMFs from lower-dimensional marginalized PMFs under certain low-rank conditions and propose a nonnegative coupled low-rank tensor factorization algorithm.

Past Projects

wcl2015 Fully-Coherent Reader with Commodity SDR for Gen2 FM0 and Computational RFID
N. Kargas, F. Mavromatis and A. Bletsas
IEEE Wireless Communications Letters, 2015
This work offers a complete SDR reader with coherent detection, exploitation of FM0 line coding memory in Gen2 tags, careful handling of symbol synchronization and implementation and testing of Gen2 in a commodity SDR, utilizing a single transceiver card.
rfidta2014 Channel Coding for Increased Range Bistatic Backscatter Radio: Experimental Results
P. N. Alevizos, N. Fasarakis-Hilliard, K. Tountas, N. Agadakos, N. Kargas and A. Bletsas
IEEE RFID Technology and Applications Conference (RFID-TA), 2014

This work offers concrete, low-complexity channel coding for the bistatic scatter radio channel. The theoretical design is experimentally validated with a commodity software-defined radio (SDR) reader.