Elsa López Pérez

I build formally verified cryptographic protocols for tiny IoT devices, train models that learn from messy real-world data, and turn wearable signals into performance insight.

PhD researcher at Inria & Sorbonne Université · Maths & Physics Bachelor · Data Scientist and ML Engineer · Level 2 triathlon coach · triathlete · saxophonist.

selected projects — pick a world

Career

Experience & Education

Inria · SorbonneNov 2023 – Dec 2026
Paris, France
· mobility @ HPI, Potsdam

Inria – Sorbonne Université

PhD Researcher — Secure IoT Protocols

Designed, implemented, and formally verified the EDHOC-PSK authentication protocol for constrained IoT devices using SAPIC+
Identified authentication and identity-protection weaknesses under advanced adversary models including post-quantum attackers
Benchmarked performance and memory footprint on ARM Cortex-M platforms
Contributed protocol specifications directly to the IETF LAKE standardization process
International research mobility at the Hasso Plattner Institute (HPI), Potsdam — developing a project on the integration of EDHOC into 5G networks (ongoing)

Nokia Bell LabsMar – Aug 2023
Paris, France

Nokia Bell Labs

Research Intern — Reinforcement Learning & Multi-Robot Systems

Integrated and re-trained a PPO-based multi-agent navigation policy within a ROS robotics stack
Led simulation-to-real transfer analysis, resolving deployment failures from sensing noise and actuation delays
Improved navigation success rate to 98% in randomised high-fidelity simulation

IBISC · Paris-SaclayMay – Aug 2022
Évry, France

IBISC – Université Paris-Saclay

ML Intern — AI for Pharmaceutical Imaging

Built a semi-supervised deep learning pipeline for automated classification of 190k+ MFI particle images
Improved binary classification accuracy from 70% to 86% by addressing class imbalance and dataset mislabelling

Education2016 – 2026

Academic Background

PhD · Master · Double Bachelor

PhD in Secure Protocols — Sorbonne Université & Inria (2023–2026)
EIT Digital Master in Data Science & AI — Université Paris-Saclay & Università di Trento (2021–2023)
Double Bachelor, Mathematics & Physics — Univ. of Valladolid & Univ. of Dundee, Erasmus (2016–2021)

Research

Selected Publications

↗ View full list on Google Scholar

2026

embedded-cal: A Formally Verified Cryptographic Provider for Embedded Platforms🏆 Best Paper Award

Elsa López Pérez et al. — SPT-IoT 2026

2026

Formal Verification of EDHOC-PSK: A Symbolic Approach

Elsa López Pérez et al. — ACM ASIACCS 2026

2025

Pre-Shared Key Authentication With EDHOC: The Security–Performance Tradeoff

Elsa López Pérez et al. — IEEE Access 2025

2024

EDHOC Authenticated with Pre-Shared Keys (PSK)

Elsa López Pérez et al. — IETF Internet Draft 2024

2024

EDHOC is a New Security Handshake Standard

Elsa López Pérez et al. — IEEE Computer 2024

Toolbox

Skills

Programming Languages

Rust (embedded cryptography), C/C++ (microcontroller drivers), Python (ML pipelines)

ML & Data Science

PyTorch, scikit-learn, TensorFlow, time-series analysis, multi-agent RL, VAEs, PPO

Formal Methods

Tamarin Prover, ProVerif, SAPIC+, symbolic security analysis, protocol modelling

Embedded & Hardware

ARM Cortex-M, STM32WBA55, nRF54, bare-metal Rust, hardware crypto accelerators

Robotics

ROS, Gazebo simulation, sim-to-real transfer, multi-robot coordination

Standardization

IETF LAKE Working Group contributor, EDHOC-PSK draft author

Languages

Spanish (native), English (proficient), French (proficient), Italian (intermediate)

Beyond the terminal

Professional saxophone diploma · Triathlon athlete

Coaching Certifications

Level 2 Triathlon Coach · Advanced Triathlon (Middle & Long Distance) · Long-Distance Triathlon Training · TrainingPeaks · HIIT Training · Cycling Biomechanics

Machine Learning, AI & Computer Vision

From multi-robot reinforcement learning to variational autoencoders on 190k pharmaceutical images — models that learn structure from messy, real, high-stakes data.

Reinforcement Learning · Robotics

Multi-Robot Navigation with RVO-Shaped RL

Curriculum training phases: 4 robots, 10 robots, 10 robots with walls

0%nav success · 20 agents

At Nokia Bell Labs: integrated and re-trained a PPO multi-agent policy in a full ROS stack, then led the sim-to-real transfer closing the Gazebo↔hardware gap with an RVO-shaped reward and a 4→10→20 robot curriculum.

PythonROSGazeboPPO

GitHub ↗

Deep Learning · Computer Vision

Semi-Supervised VAE for Pharmaceutical Imaging

Particle snippets extracted from a microscopic flow-imaging scan

0%accuracy (from 70%)

A convolutional VAE learning compact latent representations over 190k+ microscopic flow-imaging particles, leveraging a huge unlabelled pool for automated QC under class imbalance and label noise (IBISC).

PyTorchscikit-learnOpenCVt-SNE

GitHub ↗

Machine Learning · Statistical Theory

Supervised Learning with Gaussian Processes

Bachelor's thesis on the theory of GP regression — its equivalence to kernel ridge regression and PAC-Bayesian guarantees — applied to map a 21-D input to the 7 torque outputs of an anthropomorphic robotic arm.

PythonGaussian ProcessesPAC-Bayes

Full thesis ↗

Data Mining · Recommendation

Query Recommendation & Importance Mining

A two-part system: a query-recommendation engine built from user–query interaction data, plus a scoring method quantifying how important a query is to each user — applied to a real-estate search dataset.

PythonJupyterQuery Mining

GitHub ↗

High-Performance Computing

Parallel Dijkstra for HPC Clusters

Serial, OpenMP and MPI implementations of shortest-path search sharing one benchmarking pipeline, with Python graph generation and qsub cluster jobs — measuring speedup and scaling on an MPICH cluster.

CMPIOpenMPPython

GitHub ↗

Sports Performance & Physiology

Where rigorous data work meets the human body. I turn longitudinal wearable signals into recovery, load and performance insight — and coach it on the road too.

HRV

recovery signal

Sleep

architecture

Load

training stress

♥ HRV, resting HR & menstrual-cycle phases · real wearable data

Daily HRV and resting HR over three months, shaded by menstrual cycle phase

Continuous Garmin record: daily HRV (rMSSD, blue) against its baseline band and resting HR (red), with the background shaded by cycle phase — menstrual, follicular, ovulation and luteal. Click to enlarge.

Time-Series ML · Personal Research

Cycle-Phase & Ovulation Prediction from Wearables

An end-to-end privacy-preserving platform collecting longitudinal HRV, resting HR, sleep, stress and training-load data from multiple participants, with a multi-class time-series pipeline to model cycle phases and detect ovulation from continuous biosignals.

FastAPI web app + packaged cross-platform desktop app that gathers wearable data without ever storing credentials
Rolling-window feature engineering — lag features, past-only rolling means, day-to-day deltas across HRV, sleep stages & load
Statistically validated the follicular→luteal (ovulation) transition is detectable from resting HR & HRV (large effect sizes)

Key result · ovulation detection

Cycle-aligned resting HR and HRV across the estimated ovulation boundary

Averaged across 27 cycles and aligned to the estimated ovulation day (green line): resting HR steps up after ovulation while HRV drops down — a clean, statistically robust physiological signature. Click to enlarge.

RHR ↑ +1.81 bpm · d = 1.53 HRV ↓ −5.08 ms · d = −0.59 p ≤ 0.005 · n = 27 cycles

GitHub ↗

🎓 Coaching education & certifications

Advanced Triathlon Coaching

Triathlon · Middle & Long Distance

Triathlon Coaching — Long-Distance Training

Triathlon

TrainingPeaks

Endurance Sports

HIIT Training

Sports Science

Cycling Biomechanics Coaching

Cycling

🥇 Level 2 certified triathlon coach

🏊🚴🏃 Triathlon athlete

📊 Recovery · Load · Performance