Tomàs Ortega

Research Software Engineer at Princeton University, working on AI for (formalized) mathematics.

I received a PhD in Electrical Engineering and Computer Science at UC Irvine, advised by Hamid Jafarkhani. My thesis was on Collaborative Learning under Communication Constraints.

More broadly, I am interested in optimization, information theory, and AI. I also organized a group for people in the greater Los Angeles area to learn how to write mathematical proofs in the Lean formal verification language.

Education

• 2021-2026: PhD in Electrical Engineering and Computer Science, UC Irvine, Irvine.
• 2020-2021: M.S. in Mathematics, UPC, Barcelona.
• 2015-2020: B.S. in Mathematics, CFIS, UPC, Barcelona.
• 2015-2020: B.Eng. in Telecommunications Engineering, CFIS, UPC, Barcelona.

Professional and Research Experience

• Jun 2026-Present: Research Software Engineer at Princeton University. Research and tool development in AI for mathematics; Princeton.
• Jun-Aug 2025: Research Intern at Nokia Bell Labs. Designed trustless decentralized data storage and management systems; Murray Hill.
• Jun-Aug 2024: Research Intern at the Vector Institute. Investigated and improved the optimization of large language models in federated learning scenarios; Toronto.
• Jun-Aug 2022: Graduate Fellow at NASA Jet Propulsion Laboratory. Designed and supervised experimental channel-sounding tests for Lunar South Pole communications and developed novel algorithms for cooperative and compressed localization; Pasadena.
• Sep-Dec 2020: Research Assistant at UPC, Signal Theory and Communications Department. Designed and optimized 5G coverage estimators in urban scenarios, then implemented and tested the proposed solutions; Barcelona.
• Feb-Jul 2020: JVSRP Intern at NASA Jet Propulsion Laboratory. Developed and implemented an adaptive-sweep algorithm for carrier acquisition and tracking in spacecraft radios; Pasadena.
• Jan-Mar 2019: Research Project Collaborator at HP. Designed an ultrasound positioning system for mobile printers with a sensor network, specializing in position calculation and communications software; Barcelona.
• Jun-Aug 2018: Intern at BaseTIS. Enhanced task automation and parallelization for data analysis and reporting for Gas Natural Informática; Barcelona.

Skills

• Distributed systems and software engineering: decentralized optimization, consensus protocols, trustless and decentralized data ledgers, and cryptographic primitives.
• AI/ML systems: federated learning (quantized, asynchronous, and personalized), privacy-preserving aggregation, model-parallel training pipelines, and performance benchmarking.
• Programming and prototyping: Python with TensorFlow and PyTorch, C++, Java, Git, Docker, and CI/CD pipelines.
• Research and R&D deliverables: publication pipelines for IEEE and ICML conferences and journals, patent drafting, experimental design, and open-source library contributions.
• Tools and platforms: Linux, Kubernetes, AWS/GCP, LaTeX, the Lean theorem prover, Jupyter, and GitHub Actions.
• Communication and collaboration: technical leadership, conference presentations, peer review for IEEE T-COM, DCC, and IEEE SPS, and cross-disciplinary collaboration.
• Languages: Catalan and Spanish (native), English (proficient; Cambridge Proficiency certificate), and French (conversational; DELF certificate).

Publications

1. Ayush Khaitan, Liam Fowl, Tomàs Ortega, Alex Kontorovich, Sanjeev Arora. Does a Frontier Large Language Model Know What It Can Solve? A 100-Conjecture Calibration Study. In NeurIPS MATH-AI Workshop, 2026. openreview.net/forum?id=CT9WOChlRl
2. Ayush Khaitan, Liam Fowl, Tomàs Ortega, Alex Kontorovich, Sanjeev Arora. Improving semantic equivalence in research-level math formalization. In NeurIPS MATH-AI Workshop, 2026. openreview.net/forum?id=OFimGMpfHW
3. Tomàs Ortega, Simon Park, Ayush Khaitan, Liam Fowl, Alex Kontorovich, Sanjeev Arora. Do Theorem Provers Need Theorem-Proving Agents? mini-Lean-agent: A Coding-Agent Baseline for Lean. In NeurIPS MATH-AI Workshop, 2026. openreview.net/forum?id=gInXazr59f
4. Nora Bradford, Tomàs Ortega, Brian Maniscalco, Megan A. K. Peters. Illusory evidence asymmetrically impacts visual performance and metacognition. PsyArXiv preprint, 2026. doi:10.31234/osf.io/7zf4j_v1
5. Tomàs Ortega, Hamid Jafarkhani. Decentralized Parameter-Free Online Learning with Compressed Gossip. 2026. arXiv:2605.27831
6. Tomàs Ortega, Chun-Yin Huang, Xiaoxiao Li, Hamid Jafarkhani. Communication Compression for Distributed Learning with Aggregate and Server-Guided Feedback. 2025. arXiv:2512.22623
7. Tomàs Ortega, Hamid Jafarkhani. Decentralized Parameter-Free Online Learning. 2025. arXiv:2510.15644
8. Tomàs Ortega, Chun-Yin Huang, Xiaoxiao Li, Hamid Jafarkhani. Communication Compression for Distributed Learning without Control Variates. 2024. arXiv:2412.04538
9. Tomàs Ortega, Hamid Jafarkhani. Quantized and Asynchronous Federated Learning. IEEE Transactions on Communications, pp. 1-1, 2024. doi:10.1109/TCOMM.2024.3471996
10. Tomàs Ortega, Hamid Jafarkhani. Decentralized Optimization in Networks with Arbitrary Delays. In ICC 2024 - IEEE International Conference on Communications, pp. 794-799, 2024. doi:10.1109/ICC51166.2024.10622164
11. Simeon Ball, Tomàs Ortega. Practical implementation of geometric quasi-cyclic LDPC codes. 2024. arXiv:2405.20524
12. Tomàs Ortega, Hamid Jafarkhani. Decentralized Optimization in Time-Varying Networks with Arbitrary Delays. 2024. arXiv:2405.19513
13. Tomàs Ortega, Hamid Jafarkhani. Asynchronous Federated Learning with Bidirectional Quantized Communications and Buffered Aggregation. 2023 International Conference on Machine Learning, Federated Learning and Analytics in Practice Workshop, 2023. doi:10.48550/arXiv.2308.00263
14. Tomàs Ortega, Hamid Jafarkhani. Gossiped and Quantized Online Multi-Kernel Learning. IEEE Signal Processing Letters, 30, pp. 468-472, 2023. doi:10.1109/LSP.2023.3268988
15. Tomàs Ortega, Antonio Pascual-Iserte, Olga Muñoz. LOS/NLOS Estimators for mmWave Cellular Systems With Blockages. IEEE Wireless Communications Letters, 11(1), pp. 121-125, 2022. doi:10.1109/LWC.2021.3122090
16. Tomàs Ortega, Marc Sanchez Net, Kar-Ming Cheung, Dariush Divsalar. Adaptive-Sweep Algorithm for Spacecraft Carrier Acquisition and Tracking: System Analysis and Implementation. In 2021 IEEE Aerospace Conference (50100), pp. 1-9, 2021. doi:10.1109/AERO50100.2021.9438340
17. Tomàs Ortega, Marc Sanchez Net, Dariush Divsalar, Kar-Ming Cheung. Acquisition and tracking of high dynamics Doppler profiles for space applications. In 2021 IEEE Aerospace Conference (50100), pp. 1-20, 2021. doi:10.1109/AERO50100.2021.9438418

Patents and Intellectual Property

• 2024: Systems and Methods for Quantized Machine Learning, Federated Learning and Bidirectional Network Communication. Patent pending; US 2024/0354589 A1.
• 2023: Quasi-Cyclic LDPC Codes Based on Generalised Quadrangles. National phase pending; WO 2023/218050 A1.

Open-Source Software

• Capacity Atlas (capacityatlas.org; github.com/TomasOrtega/CapacityAtlas). A community-maintained registry of channel capacities, known bounds, open gaps, formally stated claims, and machine-checked Lean proofs.
• commonly-cited (web app; github.com/TomasOrtega/get-commonly-cited). A web app and command-line tool that resolves bibliographies, expands “et al.,” and identifies the most frequently cited people.
• DECO: Decentralized Coin-Betting for Online Learning (github.com/TomasOrtega/DECO). Reproduces the experiments and figures from “Decentralized Parameter-Free Online Learning” [7].
• CAFe: Compressed Aggregate Feedback (github.com/TomasOrtega/CAFe). Implements the experiments for “Communication Compression for Distributed Learning with Aggregate and Server-Guided Feedback” [6].
• FLSim: Federated Learning Simulator (github.com/TomasOrtega/FLSim). Reproduces experiments from “Quantized and Asynchronous Federated Learning” [9, 13]. Built in Python, with support for custom compression and asynchronous aggregation.
• DT-GO: Decentralized Optimization (github.com/TomasOrtega/DT-GO). Implements algorithms from “Decentralized Optimization in Time-Varying Networks with Arbitrary Delays” [12]. A Python prototype with performance benchmarks for decentralized logistic regression and federated-learning tasks.
• Noisy Lattice Denoising (github.com/TomasOrtega/Noisy-lattice-problem). A MATLAB, Python, and AMPL solver for a noisy lattice-point recovery problem.

Merits and Awards

• 2026: Henry Samueli Endowed Fellowship, Samueli School of Engineering at UCI; awarded.
• 2024: EECS Graduate Student of the Year, Engineering Student Council at UCI; awarded.
• 2023: Signal Processing Scholarship, IEEE Signal Processing Society; awarded.
• 2023: Early Career Invitation, ICML Federated Learning Workshop; awarded.
• 2023: Évariste Galois Prize for Best M.S. Thesis in Catalonia, Catalan Society of Mathematics; honorable mention.
• 2022: JPL Graduate Fellowship, NASA Jet Propulsion Laboratory; awarded.
• 2021: Electrical Engineering and Computer Science Department Fellowship, Samueli School of Engineering at UCI; awarded.
• 2021: Balsells Graduate Fellowship, Balsells Program; awarded.
• 2018-2020: Google Hash Code, 2nd, 1st, and 1st team in Spain; 171st, 75th, and 53rd globally.
• 2018: Kernel Analytics Datathon, 2nd place for classifying Parkinson’s disease symptom onset from sensor data.
• 2015: CFIS Scholarship, CFIS; awarded.

Teaching

• Jan-Mar 2026: Probability for Engineers, EECS 55, Irvine.
• Jan-Mar 2025: Probability for Engineers, EECS 55, Irvine.
• Jan-Mar 2024: Probability for Engineers, EECS 55, Irvine.

Leadership and Service

• 2023-2024: Graduate Student Representative at the UCI Council on Planning and Budget. Represented graduate students at the UCI Council on Planning and Budget, Irvine.
• 2023-2024: Graduate Student Representative at the UCI Samueli School of Engineering Graduate Studies Committee. Represented graduate students on the Graduate Studies Committee, Irvine.
• 2023-2024: Council Member for the School of Engineering at UCI’s Associated Graduate Students. Represented the School of Engineering on the elected Associated Graduate Students council, Irvine.