Cedric Renggli

cedric_renggli.jpg

I am a Senior Researcher and Lecturer at ETH’s Systems Group, working with Ana Klimovic.

I do research at the intersection of AI systems and data management. My goal is to make AI systems efficient and reliable, so that users reach validated results with less data, computation, and expert effort.

Previously, I was a Senior Researcher at Apple and a PostDoc at UZH (DaST, with Dan Olteanu). I received my PhD from ETH’s Systems Group, advised by Ce Zhang.

Research

With talented students and collaborators, I work on two broad areas around AI systems.

Efficiency

I make AI cheaper to develop, train, and serve. On the systems side, this ranges from communication-efficient distributed training (SparCML, SC’19) to my current work on efficient serving and post-training of agentic AI, together with CSCS and the Swiss AI Initiative. On the data side, I reduce the samples, labels, and data movement needed to reach a result, for example when testing models (ease.ml/ci, SysML’19), selecting pretrained models (SHiFT, VLDB’22), training inside databases (CorgiPile, VLDB Journal’24), or maintaining vector indexes (Ada-IVF).

Reliability

I study what limits the quality of AI and how to verify that it works. On the data side, this includes estimating the best achievable accuracy from the data alone (Snoopy, ICDE’23; FeeBee, NeurIPS’21), statistical guarantees for model testing (ease.ml/ci), and separating solution, data, and metric limitations in Text2SQL evaluation (2025). On the systems side, I study quality issues that arise from execution environments and from efficiency optimizations themselves, particularly in agentic workloads.

Selected Publications

  1. Fundamental Challenges in Evaluating Text2SQL Solutions and Detecting Their Limitations
    Cedric Renggli, Ihab F Ilyas, and Theodoros Rekatsinas
    arXiv preprint arXiv:2501.18197, 2025
    arXiv
  2. SHiFT: an efficient, flexible search engine for transfer learning
    Cedric Renggli, Xiaozhe Yao, Luka Kolar, and 3 more authors
    Proceedings of the VLDB Endowment, 2022
    arXiv PDF
  3. Automatic feasibility study via data quality analysis for ml: A case-study on label noise
    Cedric Renggli, Luka Rimanic, Luka Kolar, and 2 more authors
    2023 IEEE 39th International Conference on Data Engineering (ICDE), 2023
    arXiv PDF
  4. A Data Quality-Driven View of MLOps
    Cedric Renggli, Luka Rimanic, Nezihe Merve Gürel, and 3 more authors
    IEEE Data Engineering Bulletin, 2021
    arXiv PDF
  5. Continuous Integration of Machine Learning Models with ease.ML/CI: Towards a Rigorous Yet Practical Treatment
    Cedric Renggli, Bojan Karlas, Bolin Ding, and 4 more authors
    In SysML Conference, 2019
    arXiv PDF
  6. SparCML: High-performance sparse communication for machine learning
    Cedric Renggli, Saleh Ashkboos, Mehdi Aghagolzadeh, and 2 more authors
    In Proceedings of the International Conference for High Performance Computing, Networking, Storage and Analysis, 2019
    arXiv PDF
  7. Stochastic gradient descent without full data shuffle: with applications to in-database machine learning and deep learning systems
    Lijie Xu, Shuang Qiu, Binhang Yuan, and 8 more authors
    The VLDB Journal, 2024
    arXiv PDF