Hrayr Harutyunyan
Scholar

Hrayr Harutyunyan

Google Scholar ID: GaCGz8wAAAAJ
Research Scientist, Google DeepMind
Large Language ModelsPre-trainingPost-trainingLearning TheoryInformation Theory
Citations & Impact
All-time
Citations
2,703
 
H-index
9
 
i10-index
9
 
Publications
20
 
Co-authors
19
list available
Resume (English only)
Academic Achievements
  • Paper 'Mixture-of-recursions: Learning dynamic recursive depths for adaptive token-level computation' accepted at NeurIPS 2025 (Sept 2025)
  • Paper 'Supervision Complexity and its Role in Knowledge Distillation' accepted at ICLR 2023
  • Paper 'Formal limitations of sample-wise information-theoretic generalization bounds' accepted at 2022 IEEE Information Theory Workshop
  • Paper 'Failure Modes of Domain Generalization Algorithms' accepted at CVPR 2022
  • Paper 'Information-theoretic generalization bounds for black-box learning algorithms' accepted at NeurIPS 2021
  • Paper 'Estimating informativeness of samples with Smooth Unique Information' accepted at ICLR 2021
  • Paper 'Improving generalization by controlling label-noise information in neural network weights' accepted at ICML 2020
  • Paper 'Reducing overfitting by minimizing label information in weights' accepted at NeurIPS'19 Information Theory and Machine Learning Workshop
  • Paper 'Fast structure learning with modular regularization' accepted as a spotlight at NeurIPS 2019
  • Received free NeurIPS 2020 registration as a top 10% reviewer
  • Invited speaker at Rising Stars in AI Symposium 2023, KAUST, Saudi Arabia (Feb 2023)