About Me

I am a Research Scientist at Snap Research on the User Modeling and Personalization (UMaP) team led by Neil Shah. My work focuses on advancing the state of the art in machine learning for large-scale recommendation systems, and I am currently interested in generative recommendation and representation learning utilizing multimodal user interaction data.

Broadly, I study how learning systems can make effective and responsible decisions in complex environments shaped by human, social, and economic factors.

Before joining Snap, I was a Machine Learning Scientist at TikTok and I designed several components of TikTok’s recommendation system. I earned my Ph.D. in Computer Science from the Georgia Institute of Technology, advised by Prof. Jacob Abernethy and Prof. Jamie Morgenstern, where my research explored the theoretical foundations of learning systems informed by social and economic considerations, spanning topics such as mechanism design, differential privacy, robust machine learning, and active learning. I hold a Bachelor’s degree in Computer Science and Engineering from IIT Kanpur.

For more details, please see my CV.

Publications

• Sequential Data Augmentation for Generative Recommendation
  G. Lee, B. Kumar, C. M. Ju, T. Zhao, K. Shin, N. Shah, L. Collins
  WSDM 2026
  [pdf]

• Learning Universal User Representations Leveraging Cross-Domain User Intent at Snapchat
  C. M. Ju, L. Neves, B. Kumar, L. Collins, T. Zhao, Y. Qiu, Q. Dou, et al.
  SIGIR 2025
  [pdf]

• Generative Recommendation with Semantic IDs: A Practitioner’s Handbook
  C. M. Ju, L. Collins, L. Neves, B. Kumar, Y. Y. Wang, T. Zhao, N. Shah
  CIKM 2025 [Best Paper Award in Resource Papers]
  [pdf], [Library]

• Private Mechanism Design via Quantile Estimation
  Y. Yang, T. Xiao, B. Kumar, J. Morgenstern
  ICLR 2025
  [pdf], [poster]

• Revisiting self-attention for cross-domain sequential recommendation
  C. M. Ju, L. Neves, B. Kumar, L. Collins, T. Zhao, Y. Qiu, Q. Dou, S. Nizam, S. Yang, N. Shah
  KDD 2025
  [pdf]

• Accelerated Federated Optimization with Quantization
  Y. Youn, B. Kumar, J. Abernethy
  IEEE Data Engineering Bulletin 2023
  [pdf]

• ActiveHedge: Hedge meets Active Learning
  B. Kumar, J. Abernethy, V. Saligrama
  ICML 2022 (Spotlight)
  [pdf], [poster], [talk]

• Observation Free Attacks on Stochastic Bandits
  Y. Xu, B. Kumar, J. Abernethy
  NeurIPS 2021
  [pdf]

• Bridging Truthfulness and Corruption-robustness in Multi-Armed Bandit Mechanisms
  Y. Xu, B. Kumar, J. Abernethy, T. Lykouris
  Incentives in ML Workshop at ICML 2020
  [pdf], [talk]

• Learning Auctions with Robust Incentive Guarantees
  B. Kumar, J. Abernethy, R. Cummings, J. Morgenstern, S. Taggart
  NeurIPS 2019
  [pdf], [poster]

Preprint(s)

• Optimal spend rate estimation and pacing for ad campaigns with budgets
  B. Kumar, J. Morgenstern, O. Schrijvers
  arXiv (2022)
  [pdf], [talk]