Mandar H. Chandorkar

The generic can be more intense than the concrete. - Jorge Luis Borges

I’m a scientist and engineer currently based in Rotterdam, the Netherlands.

Currently, I work as a Data Scientist at Connecterra B.V, a dairy-tech scale up in Amsterdam. During my current tenure, I gained valuable experience in:

End-to-end machine learning:
- Analysis, prototying and proof of concept for new models / services.
- Building ML experimentation software stacks.
- Large scale data preparation and model evaluation pipelines, going beyond text-book metrics.
- Engineering real-time services serving ML models.
tinyML model optimisations such as quantisation, enabling power savings on edge deployments.

Previously I used to be a researcher at the CWI Amsterdam and the TAU group at the INRIA Paris-Saclay. During my PhD days, my research interests broadly covered machine learning applications in physical systems. I am interested in all the interesting problems that fall in between, such as, forecasting, simulation, uncertainty quantification, and system identification.

In my free time, I enjoy literature; fiction and (sometimes) non-fiction, and expanding my horizons with music from around the world.

news

Dec 1, 2019	Started working at Connecterra B.V
Nov 14, 2019	Defended PhD at the TU Eindhoven

selected publications

AGU

Probabilistic forecasting of the disturbance storm time index: An autoregressive Gaussian process approach

Chandorkar, M, Camporeale, E, and Wing, S

Space Weather 2017

Bib HTML

@article{chandorkar2017probabilistic,
  title = {Probabilistic forecasting of the disturbance storm time index: An autoregressive Gaussian process approach},
  author = {Chandorkar, M and Camporeale, E and Wing, S},
  journal = {Space Weather},
  volume = {15},
  number = {8},
  pages = {1004--1019},
  year = {2017},
  publisher = {American Geophysical Union (AGU)},
}

Chapter 9 - Probabilistic Forecasting of Geomagnetic Indices Using Gaussian Process Models

Chandorkar, Mandar, and Camporeale, Enrico

2018

Abs Bib HTML

In this chapter, we give the reader an in-depth view into building of probabilistic forecasting models for geomagnetic time series using the Gaussian process methodology outlined in the previous chapters. We highlight design decisions and practical issues that must be addressed in order to use Gaussian process models for probabilistic prediction of a quantity of interest. As a pedagogical example, we formulate, train, and test a family of Gaussian process auto-regressive models for 1-h ahead prediction of the Dst geomagnetic index.

@incollection{CHANDORKAR2018237,
  title = {Chapter 9 - Probabilistic Forecasting of Geomagnetic Indices Using Gaussian Process Models},
  editor = {Camporeale, Enrico and Wing, Simon and Johnson, Jay R.},
  booktitle = {Machine Learning Techniques for Space Weather},
  publisher = {Elsevier},
  pages = {237 - 258},
  year = {2018},
  isbn = {978-0-12-811788-0},
  doi = {https://doi.org/10.1016/B978-0-12-811788-0.00009-3},
  url = {https://www.sciencedirect.com/science/article/pii/B9780128117880000093},
  author = {Chandorkar, Mandar and Camporeale, Enrico},
  keywords = {Gaussian process, Space weather, Probabilistic forecasting, Geomagnetic indices},
}

ICLR

Dynamic Time Lag Regression: Predicting What & When

Chandorkar, Mandar, Furtlehner, Cyril, Poduval, Bala, Camporeale, Enrico, and Sebag, Michele

In International Conference on Learning Representations 2020

Bib HTML Poster

@inproceedings{chandorkar2019dynamic,
  title = {Dynamic Time Lag Regression: Predicting What \& When},
  author = {Chandorkar, Mandar and Furtlehner, Cyril and Poduval, Bala and Camporeale, Enrico and Sebag, Michele},
  booktitle = {International Conference on Learning Representations},
  year = {2020},
}