# Kaushal Paneri — full content bundle

> Senior applied scientist at Microsoft, leading large-scale ML systems in auction and autobidding at Bing Ads. Work at the intersection of causal inference, reinforcement learning, and foundation models for large-scale recommender systems. 16 peer-reviewed publications at NeurIPS, AAAI, IEEE TBD, ACM RecSys. 2 granted US patents. Visiting Lecturer at Northeastern University (2021-2023).

## About

Kaushal Paneri is a senior applied scientist at Microsoft, where he leads large-scale ML systems in auction and autobidding at Bing Ads. His work sits at the intersection of causal inference, reinforcement learning, and foundation models — figuring out how to safely improve systems that can only observe the consequences of their own past decisions.

He completed his MS in Data Science (thesis track) at Northeastern University under Robert Ness, Jan-Willem van de Meent, and Olga Vitek, where his thesis on integrating Markov processes with structural causal models became a NeurIPS 2019 paper. He taught graduate machine learning (CS 6140) at Khoury College as a visiting lecturer from 2022–2024. Before Northeastern, he spent three years at TCS Research Lab in Delhi doing applied ML research and publishing in top-tier ML conferences, producing 7 publications and 2 US patents.

## Research Areas

- Reinforcement learning and bandit algorithms for auction tuning and autobidding
- Foundation models and world models for marketplace forecasting (Chronos)
- Causal inference and counterfactual evaluation for recommender systems
- Off-policy estimation in high-dimensional action spaces (SigIS method)
- Structural causal models and Bayesian inference

## Career

### Senior Data & Applied Scientist · Microsoft (2024 – present)
Foundation-model forecasting (Chronos), off-policy estimation research for high-dimensional action spaces, model predictive control for marketplace optimization. Program committee member for CONSEQUENCES '25 at ACM RecSys.

### Data & Applied Scientist → DAS2 · Microsoft (2019 – 2024)
Sole research owner of Microsoft's click prediction ML stack. Transitioned the system from a legacy Probit model to a Random Forest / ONNX architecture. Developed simulator-guided importance sampling methods (SigIS, first-author oral at RecSys '24). Created the counterfactual evaluation backbone for Microsoft's autobidding platform. Mentored three research internships, two of which produced production systems and one a peer-reviewed publication.

### Visiting Lecturer · Northeastern University (2021 – 2023)
Khoury College of Computer Sciences. Taught DS 5220: Supervised Machine Learning and Learning Theory (graduate, two offerings) and DS 4400: Machine Learning and Data Mining 1 (undergraduate).

### MS, Data Science (Thesis Track) · Northeastern University (2017 – 2019)
Thesis on integrating Markov processes with structural causal models for counterfactual inference. Advisors: Olga Vitek, Robert Ness, Jan-Willem van de Meent. Published at NeurIPS 2019. GPA 3.78.

### Researcher · TCS Research Lab, Delhi (2014 – 2017)
Designed and built a Bayesian Network Inference Engine deployed across 20+ client organizations. Published 7 papers at IEEE VIS, VAST, FUSION, and ICDMW. Won 4 international data challenges. Research produced 2 granted US patents (US 10,430,417, US 10,013,634).

### B.Tech., Computer Engineering · Charusat University (2010 – 2014)
Gujarat, India. TCS CodeVita national coding competition, top 0.006%.

## Publications

- Combining Open-box Simulation and Importance Sampling for Tuning Large-Scale Recommenders. K. Paneri, M. Munje, K. Singh Maurya, et al.. CONSEQUENCES @ ACM RecSys 2024. https://arxiv.org/abs/2410.03697
- Adaptive Mixture Importance Sampling for Automated Ads Auction Tuning. Y. Jia, K. Paneri, R. Huang, et al.. CONSEQUENCES @ ACM RecSys 2024. https://arxiv.org/abs/2409.13655
- Leveraging Structured Biological Knowledge for Counterfactual Inference: A Case Study of Viral Pathogenesis. J. Zucker, K. Paneri (co-first), S. Mohammad-Taheri, et al.. IEEE Transactions on Big Data, 7(1), 25–37, 2021.
- Integrating Markov Processes with Structural Causal Modeling Enables Counterfactual Inference in Complex Systems. R. Ness, K. Paneri, O. Vitek. NeurIPS 2019. https://proceedings.neurips.cc/paper/2019/hash/2d44e06a7038f2dd98f0f54c4be35e22-Abstract.html
- Regularizing Fully Convolutional Networks for Time Series Classification by Decorrelating Filters. K. Paneri, T.V. Vishnu, P. Malhotra, et al.. AAAI 2019, 33(01).
- Crop Planning Using Stochastic Visual Optimization. G. Sehgal, B. Gupta, K. Paneri, et al.. IEEE VDS 2017, 47–51.
- Visual Bayesian Fusion to Navigate a Data Lake. K. Singh, K. Paneri, A. Pandey, et al.. IEEE FUSION 2016, 987–994.
- Multi-Sensor Visual Analytics Supported by Machine-Learning Models. G. Sharma, G. Shroff, A. Pandey, K. Paneri, et al.. IEEE ICDMW 2015, 668–674.

## Writing

### Hello, world
*2026-04-19*

This is the first post on the rebuilt kaushal.us. I'll be writing here about
causal inference, reinforcement learning, foundation models for marketplace
forecasting, and the occasional dispatch from teaching graduate ML.

If you want to follow along, the [RSS feed](/feed.xml) is the canonical channel.