Scholar
Sasha Doubov
Google Scholar ID: E39noh8AAAAJ
Google DeepMind
Citations & Impact
Citations
160
H-index
5
i10-index
4
Publications
10
Co-authors
0
Publications
Resume (English only)
Academic Achievements
- ICML 2024: 'Beyond Chinchilla-Optimal: Accounting for Inference in Language Model Scaling Laws' – extends Chinchilla scaling laws to incorporate inference demand.
- NeurIPS 2021: 'Scalable Neural Data Server: A Data Recommender for Transfer Learning' – scalable system for recommending pre-training data across domains (e.g., medical, satellite).
- IROS 2020 (Best Application Paper Finalist): 'Pit30M: A Benchmark for Global Localization in the Age of Self-Driving Cars' – introduced large-scale LiDAR-based localization dataset.
- NeurIPS 2024 Workshop: 'Sparse Upcycling: Inference Inefficient Finetuning' – analyzes trade-offs between model quality and inference efficiency in LLM upcycling.
- NeurIPS 2022 Workshop: Investigated scalability and limitations of influence estimation in deep learning, showing many models are needed for reliable estimates.
- NeurIPS Meta-Learning Workshop 2021: Studied interaction between BatchNorm’s implicit learning rate decay and meta-learning adaptation (e.g., MAML).
Research Experience
- Research Scientist at Databricks/MosaicML (Aug 2023–present): Member of the LLM pre-training team.
- Research Scientist Intern at MosaicML (April 2023–Aug 2023): Worked on hyperparameter tuning for LLMs and training/evaluating domain-specific models.
- ML Intern (Model Efficiency Team) at Cohere (Oct 2022–): Focused on structured pruning to accelerate LLM inference and training.
- Research Intern (ML Algorithms) at Cerebras Systems (April 2022–Aug 2022): Researched unstructured sparsity for accelerating neural network training.
- Research Intern at Uber ATG (Jan–July 2019; Sept–Dec 2019): Worked on retrieval-based localization for self-driving cars.
Co-authors