Puoya graudated with a PhD from UIUC in 2021. He is now a PostDoctoral researcher at University of California, San Diego.

For more information, please visit my personal webpage and Google Scholar.

 

I was a PhD candidate in ECE at University of Illinois at Urbana-Champaing, expecting to graduate in 2021. I have been extremely fortunate to work with Prof. Dokmanic and Prof. Milenkovic. 

My research focuses on understanding the geometry of complex data graphs, and leverage it to improve the performance of common machine learning tasks, like classification and supervised alignment. 

Particularly, we proposed a semidefinite relaxation to embed tree-like structures in hyperbolic spaces from noisy, and incomplete data (pairwise comparisons and/or distances). The choice of embedding space is motivated by the fact that hyperbolic spaces are better suited (compared to Euclidean spaces) to represent weighted trees or hierarchies.

Additionally, I was working on developing a data mining tool to uncover the implicit geometry of entites given only their pairwise similarity measurements. 

Research interests

  • Geometry and Machine Learning

  • Statistical Learning Theory

  • Hyperbolic Distance Geometry Problems 

Publications

2020

Geometry of Comparisons
and
arXiv preprint arXiv:2006.09858, 2020
arXiv 
@article{tabaghi2020geometry,
  title={Geometry of Comparisons},
  author={Tabaghi, Puoya and Dokmani{\'c}, Ivan},
  journal={arXiv preprint arXiv:2006.09858},
  year={2020}
}
Hyperbolic distance matrices
and
Proceedings of the 26th ACM SIGKDD International Conference on Knowledge Discovery & Data Mining, 2020
 
@inproceedings{tabaghi2020hyperbolic,
  title={Hyperbolic distance matrices},
  author={Tabaghi, Puoya and Dokmani{\'c}, Ivan},
  booktitle={Proceedings of the 26th ACM SIGKDD International Conference on Knowledge Discovery \& Data Mining},
  pages={1728--1738},
  year={2020}
}

2019

Kinetic Euclidean distance matrices
, and
IEEE Transactions on Signal Processing, 2019
 
@article{tabaghi2019kinetic,
  title={Kinetic Euclidean distance matrices},
  author={Tabaghi, Puoya and Dokmani{\'c}, Ivan and Vetterli, Martin},
  journal={IEEE Transactions on Signal Processing},
  volume={68},
  pages={452--465},
  year={2019},
  publisher={IEEE}
}
On the move: Localization with kinetic Euclidean distance matrices
, and
ICASSP 2019-2019 IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP), 2019
 
@inproceedings{tabaghi2019move,
  title={On the move: Localization with kinetic Euclidean distance matrices},
  author={Tabaghi, Puoya and Dokmani{\'c}, Ivan and Vetterli, Martin},
  booktitle={ICASSP 2019-2019 IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP)},
  pages={4893--4897},
  year={2019},
  organization={IEEE}
}