Sami Alabed

Sami Alabed

Senior Research Scientist at Google DeepMind

Sami Alabed

I am a Senior Research Scientist at Google DeepMind, Technical Infrastructure Research group, where I specialize in the hardware-software co-design of Machine Learning systems and Silicon Architecture.

Currently, my research interest is in accelerating TPU tape-out cycles and architect novel hardware units. My expertise spans the full stack—from AI model design and performance optimization to compiler stacks and hardware synthesis. Previously, I led the integration of MCTS-driven search within production compiler stacks (XLA). My work on automatic partitioning for massive-scale models consistently outperformed expert-hand-tuned configurations, directly enhancing the efficiency of large-model training.

Beyond the lab, I am the co-founder and co-director of MenaML, a machine learning school focused on building the AI talent pipeline in the MENA region, with sucessful track record and high calibare candidates. MenaML itself is a technical non-profit organization of 30+ staff from top AI labs driven by the mission of improving science and education access in the region.

I hold a PhD in Computer Science from the University of Cambridge, Systems Research Group, where I was supervised by Dr Eiko Yoneki. My doctoral thesis focused on optimizing complex systems by leveraging causal latent structures, culminating in a custom-built, high-efficiency Structured Bayesian Optimization engine designed for large-scale graphs with a minimal memory footprint. I was also a Doctoral Student at The Alan Turing Institute, the national AI and data science institute in the UK.

I earned my MPhil Advanced Computer Science with Distinction from the University of Cambridge, supported by the Cambridge Trust and the Students of Cambridge awards. I hold a BSc Computer Science with Industrial Experience from The University of Manchester , where I received multiple awards for my contributions to student employability and departmental life, co-founded Manchester’s HackSoc, and GreatUnihack one of the longest running hackathons in the UK focused on collaboration and healthy creative competition.

My professional background includes engineering roles at Amazon, Google, Twitter, and DeepMind , focusing on large-scale distributed systems.

Latest publications

  • ASPLOS25 - Alabed, Sami, Daniel Belov, Bart Chrzaszcz, Juliana Franco, Dominik Grewe, Dougal Maclaurin, James Molloy et al. “Partir: Composing spmd partitioning strategies for machine learning.” In Proceedings of the 30th ACM International Conference on Architectural Support for Programming Languages and Operating Systems, Volume 1, pp. 794-810. 2025.
  • In review - Alabed, Sami, Dominik Grewe, Norman Alexander Rink, Masha Samsikova, Timur Sitdikov, Agnieszka Swietlik, Dimitrios Vytiniotis, and Daniel Belov. “TOAST: Fast and scalable auto-partitioning based on principled static analysis.” arXiv preprint arXiv:2508.15010 (2025).
  • NeurIPS22 MLForSystems - Alabed, S., Grewe, D., Franco, J., Chrzaszcz, B., Natan, T., Norman, T., Rink, N.A., Vytiniotis, D. and Schaarschmidt, M., 2022. Automatic Discovery of Composite SPMD Partitioning Strategies in PartIR. arXiv preprint arXiv:2210.06352. Preprint.
  • EuroSys22 MLSystems - Alabed, Sami, and Eiko Yoneki. “BoGraph: structured bayesian optimization from logs for expensive systems with many parameters.” In Proceedings of the 2nd European Workshop on Machine Learning and Systems, pp. 45-53. 2022. Preprint, Source Code, Slides, Poster, Recording.
  • EuroSys21 MLSystems - Alabed, Sami, and Eiko Yoneki. “High-Dimensional Bayesian Optimization with Multi-Task Learning for RocksDB.” In Proceedings of the 1st Workshop on Machine Learning and Systems, pp. 111-119. 2021. Proceeding, Slides, Recording.
  • MPhil Thesis - Alabed, Sami. “RLCache: Automated Cache Management Using Reinforcement Learning.” arXiv preprint arXiv:1909.13839 (2019). Thesis, Source Code.

Technical Program Committee

Previous supervisions

  • Zak Singh, Deep Reinforcement Learning for Equality Saturation. Source code.
  • Sean Parker, RLFlow: Optimising Neural Network Subgraph Transformation with World Models. Preprint.
  • Ross Tooley, Auto-tuning Spark with Bayesian Optimization.