Shreyas Kalvankar

My research interests lie in the theoretical foundations of machine learning, with a focus on understanding neural networks and the mechanisms behind their empirical success. I aim to use tools from functional analysis, operator theory, and optimization dynamics to study gradient-based learning algorithms and the structures they implicitly exploit. I am particularly interested in the mathematical mechanisms underlying training dynamics, implicit bias, generalization, and the transition between feature-learning and lazy regimes.

Broadly, I am motivated by questions at the intersection of machine learning theory and mathematical analysis, aiming to develop a clearer understanding of how modern learning algorithms behave and why they work.

I graduated with a Bachelor’s in Computer Engineering and briefly worked as an ML researcher at Relfor Labs, where I specialized in developing deep neural networks for audio data classification. I worked as a software developer at Dalton Maag where I worked on genetic algorithms.

In my free time, I enjoy exploring new languages, staying informed about technological advancements in computer science, physics, and mathematics, while also indulging in the study of history and linguistic anthropology. I also enjoy typography and sketching. I play basketball. Go Lakers!

news

Oct 10, 2025	Began my master’s thesis on Finite-width training dynamics and behaviour of large width neural networks with Prof. David Tax, Prof. Francesca Bartolucci, and Prof. Alexander Heinlein.
Sep 1, 2024	Began my master’s in Computer Science at De Technische Universiteit Delft.
Jun 17, 2022	Our work in Astronomical Image Colorization and up-scaling using Conditional Generative Adversarial Networks has been accepted at INFORMATIK 2022’s workshop ml.astro conducted by TU Dortmund in Hamburg.
Nov 22, 2021	Started working as a Software Developer at Dalton Maag, London.
Aug 14, 2021	Joined as a Machine Learning Engineer at Relfor Labs.

selected publications

EinsteinPy: A Community Python Package for General Relativity

Shreyas Bapat, Ritwik Saha, Bhavya Bhatt, and 46 more authors

2020

Abs Bib HTML

EinsteinPy performs General Relativity and gravitational physics tasks, including geodesics plotting for Schwarzschild, Kerr and Kerr Newman space-time models, calculation of Schwarzschild radius, and calculation of event horizon and ergosphere for Kerr space-time. It can perform symbolic manipulations of various tensors such as Metric, Riemann, Ricci and Christoffel symbols. EinsteinPy also features hypersurface embedding of Schwarzschild space-time, and includes other utilities and functions. It is a community-developed package and is written in Python.

@misc{bapat2020einsteinpy,
  title = {EinsteinPy: A Community Python Package for General Relativity},
  author = {Bapat, Shreyas and Saha, Ritwik and Bhatt, Bhavya and Jain, Shilpi and Jain, Akshita and Vela, Sofía Ortín and Khandelwal, Priyanshu and Shivottam, Jyotirmaya and Ma, Jialin and Ng, Gim Seng and Kerhalkar, Pratyush and Sarode, Hrishikesh Sudam and Sharma, Rishi and Gupta, Manvi and Gupta, Divya and Tyagi, Tushar and Rustagi, Tanmay and Singh, Varun and Bansal, Saurabh and Tayal, Naman and Manhas, Abhijeet and Reyna, Raphael and Kumar, Gaurav and Dixit, Govind and Kumar, Ratin and Mishra, Sashank and Jamgade, Alpesh and Singh, Raahul and Sanjay, Rohit and Shaikh, Khalid and Vidyarthi, Bhavam and K, Shamanth R Nayak and Gandham, Vineet and Vashistha, Nimesh and Das, Arnav and Saurabh and Kalvankar, Shreyas and Tarone, Ganesh and Mangat, Atul and Garg, Suyog and Gautam, Bibek and Srinivasan, Sitara and Gautam, Aayush and Singh, Swaastick Kumar and Salampuria, Suyash and Yauney, Zac and Gupte, Nihar and Shenoy, Gagan and Chan, Micky Yun},
  year = {2020},
  eprint = {2005.11288},
  archiveprefix = {arXiv},
  primaryclass = {gr-qc},
}

Astronomical Image Colorization and Up-scaling with Conditional Generative Adversarial Networks

Shreyas Kalvankar, Hrushikesh Pandit, Pranav Parwate, and 2 more authors

2022

Abs Bib HTML

This research aims to provide an automated approach for the problem of Image colorization and Single Image Super Resolution by focusing on a very specific domain: astronomical images, using Generative Adversarial Networks. We explore the usage of various models in RBG and L*a*b color spaces. We use transfer learning owing to a small data set, using pre-trained ResNet-18 as a backbone encoder and fine-tune it further. The model produces visually appealing images that are high resolution and colorized. We present our results by evaluating the GANs quantitatively using distance metrics such as L1 distance and L2 distance in each of the color spaces across all channels to provide a comparative analysis. We use Fréchet inception distance (FID) to compare the distribution of the generated images and real image to assess the model’s performance.
@article{kalvankar2022astronomical, title = {Astronomical Image Colorization and Up-scaling with Conditional Generative Adversarial Networks}, author = {Kalvankar, Shreyas and Pandit, Hrushikesh and Parwate, Pranav and Patil, Atharva and Kamalapur, Snehal}, year = {2022}, publisher = {Gesellschaft f{\"u}r Informatik, Bonn}, }