Sanit Gupta
atpugtinass@gmail.com  unscramble

About Me

I'm a quantitative researcher at AlphaGrep in London, working across strategy research, live systematic trading, and the infrastructure and machine learning models that support it. Previously, I traded options at Optiver in Amsterdam.

I studied Mechanical Engineering and Computer Science at the Indian Institute of Technology, Bombay, where I worked on problems in Multi-Armed Bandits and Reinforcement Learning with Prof. Shivaram Kalyanakrishnan and high-performance distributed computing with Prof. Shivasubramanian Gopalakrishnan. In 2019 I interned at the Max Planck Institute for Intelligent Systems under Dr. Falk Lieder, working on metacognitive reinforcement learning models of how people learn to plan.

Updates

  • [Apr 2022]  Our paper on PAC Mode Estimation accepted at AISTATS 2022
  • [Jul 2020]  Our work on modelling the spread of COVID-19 in India is now out on arXiv
  • [Jun 2020]  I was accepted into the DLRL Summer School 2020 hosted by CIFAR and Mila
  • [Aug 2019]  I was awarded the Undergraduate Research Award at IIT Bombay
  • [Jul 2019]  Our paper on reverse engineering human planning accepted at CCN 2019
  • [May 2019]  I'll be interning at the Max Planck Institute for Intelligent Systems over Summer 2019
  • [Sep 2017]  Ranked 1st in IIT Bombay in American Express's data science competition
Projects
An India-specific Compartmental Model for Covid-19: Projections and Intervention Strategies by Incorporating Geographical, Infrastructural and Response Heterogeneity

Sanit Gupta, Sahil Shah, Sumit Chaturvedi, Pranav Thakkar, Parvinder Solanki, Soham Dibyachintan, Sandeepan Roy, M. B. Sushma, Adwait Godbole, Noufal Jaseem, Pradumn Kumar, Sucheta Ravikanti, Aritra Das, Giridhara R. Babu, Tarun Bhatnagar, Avijit Maji, Mithun K. Mitra, Sai Vinjanampathy

report // code

PAC Mode Estimation using PPR Martingale Confidence Sequences

S. Jain, R. Shah, S. Gupta, D. Mehta, I. Nair, J. Vora, S. Khyalia, S. Das, V. Ribeiro, S. Kalyanakrishnan
International Conference on Artificial Intelligence and Statistics (AISTATS) 2022

paper

Multiplayer Skill Rating System

Designed the Elo-based skill rating system for Brändi Dog Online, the digital adaptation of Switzerland's best-selling board game. The system governs rating updates across solo and team modes and has been validated across 500,000+ games played by active users.
How do people learn how to plan?

Yash Raj Jain, Sanit Gupta, Vasundhara Rakesh, Peter Dayan, Frederick Callaway, Falk Lieder
Cognitive Computational Neuroscience (CCN) 2019

paper

Developing and Analyzing Algorithms for the Multi-Armed Bandit
With Prof. Shivaram Kalyanakrishnan

Developed Persistence - a generally applicable modification to multi-armed bandit algorithms. Empirically outperformed the corresponding unmodified baselines across all proposed algorithms, and proved the improvement theoretically for the ε-greedy algorithm.

report // slides

Parallel Computing for the Laplace Equation on Unstructured Grids
With Prof. Shivasubramanian Gopalakrishnan

Developed an end-to-end parallel solver for the Laplace equation over arbitrary domain shapes, with applications in thermodynamics and electrostatics. Compared naive geometric partitioning with graph-based min-cut partitioning and showed that min-cut partitioning significantly reduces inter-processor communication volume, yielding up to ~14x speedup over the serial solver.