Manos Theodosis
PhD candidate, Computer Science
CRISP Lab, Harvard University
etheodosis@g.harvard.edu
Curriculum Vitae (Updated Mar 2021)

   

I am a second year PhD student in Computer Science at Harvard University, advised by Demba Ba. I finished my undergraduate studies at the National Technical University of Athens in Greece. My thesis was conducted at the CVSP lab under the supervision of Petros Maragos, and was focused on using tropical geometry to model and analyse Weighted Finite State Transducers' algorithms. It is available here (in Greek).

My current research focuses on the unfolding of optimization algorithms into neural networks (an area of work first introduced in the seminal paper by Karol Gregor and Yann LeCun). Within that context, and assuming a group sparse generative model, I am working on proving that a neural network with a group sparse proximal operator will converge to a neighborhood of the generating matrix. My previous work focused on applying ideas from a niche field of mathematics, tropical geometry, to problems in Computer Science. In short, I worked on the applications to Weighted Finite State Transducers, curve fitting, and the analysis of the general theoretical framework.

Broadly, my research interests revolve mainly around deep learning, mathematics, and optimization. I am particularly interested in generalization and learnability analyses in deep learning, geometric and topological approaches to machine learning, and the dynamics of optimization algorithms.

I collect quotes, enjoy video games, national parks, and TV shows and movies (you can find me on trakt).


News

May 2020
Mar 2020
Feb 2020
Jan 2020
Aug 2019

Received the Gerondelis Foundation Scholarship.
Received the Robert L. Wallace Prize Fellowship.
Served as a judge at MakeHarvard 2020.
Two papers got accepted at ICASSP2020.
Started PhD in Computer Science at Harvard University.


Conference proceedings

  1. Maragos P., Theodosis E.
    "Multivariate tropical regression and piecewise-linear surface fitting"
    International Conference on Acoustics, Speech, and Signal Processing, 2020
  2. Retsinas G., Filntisis P., Efthymiou N., Theodosis E., Zlatintsi A., Maragos P.
    "Person identification using deep convolutional neural networks on short-term signals from wearable sensors"
    International Conference on Acoustics, Speech, and Signal Processing, 2020
  3. Theodosis E., Maragos P.
    "Tropical modeling of weighted transducers algorithms on graphs"
    International Conference on Acoustics, Speech, and Signal Processing, 2019
  4. Theodosis E., Maragos P.
    "Analysis of the viterbi algorithm using tropical algebra and geometry"
    International Workshop on Signal Processing Advances in Wireless Communications, 2018
Journal articles
  1. Maragos P., Charisopoulos V., Theodosis E.
    "Tropical Geometry and Machine Learning"
    Proceedings of the IEEE, vol. 109, no. 5, pp. 728-755, 2021
Book chapters
  1. Maragos P, Theodosis E.
    "Tropical Geometry and Piecewise-Linear Approximation of Curves and Surfaces on Weighted Lattices"
    In Shape Analysis: Euclidean, Discrete, and Algebraic Geometric Methods, edited by M. Breuss, A. Bruckstein, C. Kiselman, and P. Maragos, Springer, to appear
Preprints
  1. Theodosis E., Tolooshams B., Tankala P., Tasissa A., Ba D.
    "On the convergence of group-sparse autoencoders"
    arXiv, 2021
  2. Tasissa A., Theodosis E., Tolooshams B., Ba D.
    "Towards improving discriminative reconstruction via simultaneous dense and sparse coding"
    arXiv, 2020
  3. Theodosis E., Maragos P.
    "A robust adaptive pruning algorithm based on tropical geometry"
    arXiv, 2019


Talks

  1. Tropical mathematics (and why you should care)
    CRISP lab, 2019