Aravind Gollakota

I am an ML research engineer at Apple, working broadly on principled methods for reliable machine learning, including uncertainty quantification and related topics. My background is in machine learning theory and theoretical computer science. I received my PhD in CS at UT Austin in 2023 under Adam Klivans, and my undergraduate degree in math and CS at Cornell University in 2017.

Email, LinkedIn

Papers

(authors in alphabetical order unless indicated by *)

• Trace Length is a Simple Uncertainty Signal in Reasoning Models
  Siddartha Devic, Charlotte Peale, Arwen Bradley, Sinead Williamson, Preetum Nakkiran, Aravind Gollakota (*)
  Preprint, 2025
  [arxiv]
• When does a predictor know its own loss?
  Aravind Gollakota, Parikshit Gopalan, Aayush Karan, Charlotte Peale, Udi Wieder
  Foundations of Responsible Computing (FORC) 2025
  [arxiv]
• Provable Uncertainty Decomposition via Higher-Order Calibration
  Gustaf Ahdritz, Aravind Gollakota, Parikshit Gopalan, Charlotte Peale, Udi Wieder
  International Conference on Learning Representations (ICLR) 2025 (Spotlight presentation)
  [arxiv]
• An Efficient Tester-Learner for Halfspaces
  Aravind Gollakota, Adam R. Klivans, Konstantinos Stavropoulos, Arsen Vasilyan
  International Conference on Learning Representations (ICLR) 2024
  [arxiv]
• Agnostically Learning Single-Index Models using Omnipredictors
  Aravind Gollakota, Parikshit Gopalan, Adam R. Klivans, Konstantinos Stavropoulos
  Neural Information Processing Systems (NeurIPS) 2023
  [arxiv]
• Ambient Diffusion: Learning Clean Distributions from Corrupted Data
  Giannis Daras, Kulin Shah, Yuval Dagan, Aravind Gollakota, Alexandros G. Dimakis, Adam R. Klivans (*)
  Neural Information Processing Systems (NeurIPS) 2023
  [arxiv]
• Tester-Learners for Halfspaces: Universal Algorithms
  Aravind Gollakota, Adam R. Klivans, Konstantinos Stavropoulos, Arsen Vasilyan
  Neural Information Processing Systems (NeurIPS) 2023 (Oral presentation)
  [arxiv]
• A Moment-Matching Approach to Testable Learning and a New Characterization of Rademacher Complexity
  Aravind Gollakota, Adam R. Klivans, Pravesh K. Kothari
  Symposium on Theory of Computing (STOC) 2023 (invited to SIAM Journal of Computing special issue)
  [arxiv] [video]
• Hardness of Noise-Free Learning for Two-Hidden-Layer Neural Networks
  Sitan Chen, Aravind Gollakota, Adam R. Klivans, Raghu Meka
  Neural Information Processing Systems (NeurIPS) 2022 (Oral presentation)
  [arxiv] [video]
• On the Hardness of PAC-learning Stabilizer States with Noise
  Aravind Gollakota, Daniel Liang
  Quantum 6, 2022
  [arxiv]
• The Polynomial Method is Universal for Distribution-Free Correlational SQ Learning
  Aravind Gollakota, Sushrut Karmalkar, Adam R. Klivans
  Technical note, 2020
  [arxiv]
• Statistical-Query Lower Bounds via Functional Gradients
  Surbhi Goel, Aravind Gollakota, Adam R. Klivans
  Neural Information Processing Systems (NeurIPS) 2020
  [arxiv] [video]
• Superpolynomial Lower Bounds for Learning One-Layer Neural Networks using Gradient Descent
  Surbhi Goel, Aravind Gollakota, Zhihan Jin, Sushrut Karmalkar, Adam R. Klivans
  International Conference on Machine Learning (ICML) 2020
  [arxiv] [video]
• Packing Tree Degree Sequences
  Aravind Gollakota, Will Hardt, István Miklós
  Graphs and Combinatorics 36, 2020
  [arxiv]