Dr. Naila Ravelo

Biography

Naila Ravelo joined Veyra Institute in 2016 after completing a postdoctoral fellowship at the Ferric Institute of Computational Research, where she worked on asynchronous optimization methods for large-scale neural networks. She received her PhD in machine learning from the University of Calmoor in 2013, with a dissertation on convergence guarantees for distributed stochastic gradient descent under non-IID data distributions. Her undergraduate training in mathematics and statistics was at the Alderton Technical College, from which she graduated with distinction in 2008.

At Veyra, Ravelo established the Distributed Learning Systems Group to address a core practical challenge: making powerful machine learning models deployable and trainable at the edge — on devices and nodes with constrained bandwidth, intermittent connectivity, and heterogeneous compute. Her group's contributions span theory (convergence analysis under partial participation and Byzantine faults), algorithms (novel aggregation and compression schemes), and systems (a reference implementation, MeshLearn, used in three ongoing industry collaborations). She holds two patents on secure gradient aggregation protocols and has advised the Institute's commercial software-AI services team on client engagements involving edge-deployed inference.

Ravelo serves on the scientific advisory committee of the Veyra Graduate School and co-chairs the Institute's responsible AI working group. She has been principal investigator on four externally funded research programs totalling over 6.4 million cr in direct costs, and currently supervises five doctoral students and two postdoctoral researchers. She lectures the graduate course Foundations of Distributed Optimization each spring semester and contributes to the Institute's professional short course on MLOps.

Research interests

Selected publications

1. Ravelo N, Kessler M, Abara T. "Partial-participation convergence bounds for federated averaging with gradient clipping." Transactions on Distributed Machine Learning, 14(3): 211–229, 2024. VEYRA-DOI: 10.veyra/VX-2401
2. Ravelo N, Osei-Brempong P, Lindman C. "MeshLearn: a reference system for large-scale edge-federated training." Proceedings of the Calmoor Conference on Systems and ML (CCSML), pp. 88–101, 2023. VEYRA-DOI: 10.veyra/VX-2302
3. Thorvaldsen R, Ravelo N. "Secure gradient aggregation via threshold homomorphic encryption in bandwidth-limited federations." Journal of Cryptographic Engineering and Applied Systems, 9(1): 45–60, 2023. VEYRA-DOI: 10.veyra/VX-2303
4. Ravelo N. "Asynchronous SGD on non-IID partitions: tight lower bounds and an adaptive scheduler." Annals of Computational Optimization, 22: 180–198, 2021. VEYRA-DOI: 10.veyra/VX-2108
5. Ravelo N, Auzou G, Steiner R. "Topology-aware model compression for wireless federated networks." IEEE Transactions on Neural Networks and Distributed Systems, 33(7): 3120–3135, 2020. VEYRA-DOI: 10.veyra/VX-2005
6. Ravelo N, Chambers K. "Convergence of distributed gradient methods with delayed communication." Proceedings of the 10th Symposium on Parallel Learning Systems, pp. 14–26, 2018. VEYRA-DOI: 10.veyra/VX-1807

Current group members

Postdoctoral researchers

• Dr. Priya Osei-Brempong — federated optimization, Byzantine-robust aggregation
• Dr. Callum Lindman — edge hardware co-design for ML workloads

Doctoral students

• Amara Kessler — communication-efficient federated learning (Year 3)
• Yusuf Abara — non-IID robustness and distribution shift (Year 3)
• Saskia Velt — privacy amplification by subsampling in federated rounds (Year 2)
• Jonah Mirfeld — topology-aware aggregation for mesh networks (Year 1)
• Rina Tasuda — asynchronous distributed optimization under staleness (Year 1)