← back

Curriculum Vitæ

Samy-Melwan Vilhes · Ph.D student · LITIS Lab, INSA Rouen

Education

Ph.D in Time Series Anomaly Detection · LITIS Lab, INSA Rouen

Present — expected

Deep learning for time series: anomaly detection, forecasting, and foundation models for zero-shot forecasting.

Teaching Assistant — Data Analysis, INSA Rouen.

Unsupervised learning, Principal Component Analysis and Correspondence Analysis; designed and evaluated student projects in data analysis and signal processing.

M.S. in Mathematics & Artificial Intelligence · Paris-Saclay University

2022 — 2024

Jointly run by Paris-Saclay (Orsay Mathematics) and CentraleSupélec, with support from SaclAI-school.

Machine Learning (supervised, unsupervised, deep, reinforcement), Statistical Learning, High-Dimensional Modeling, Graphical Models, Optimization, Big Data Systems (SQL, HDFS), Online Learning, Conformal Prediction, Data Challenge.

B.S. in Mathematics · Paris-Saclay University

2019 — 2022

Work Experience

A.I. Research Intern · THALES

2024

Training-Free Metric for Neural Architecture Search. Explored the Neural Tangent Kernel as a training-free metric for NAS, and compared search strategies to optimize network performance.

A.I. Engineer Intern · AZAP

2023

Uncertainty Analysis in a Machine-Learning Forecasting Approach. Built a Gradient Boosting model to predict sales of promoted products, with emphasis on uncertainty analysis to improve forecast reliability.

Private Mathematics Tutor · France

2021 — 2025

Personalized mathematics tuition, helping students strengthen understanding and improve academic performance.

Selected Publications

Does Normalization Choice Matter for Causal Large Time-Series Models?

2026

ICLR 2026 Workshop (TSALM) — Spotlight · arXiv · code

PatchTrAD: A Patch-Based Transformer for Time Series Anomaly Detection

2025

EUSIPCO 2025 · arXiv · code

Understanding Neural Tangent Kernel: Key Theories and Experimental Insights

2024

Open Science · HAL · code

Distinctions & Activities

  • ICLR 2026 TSALM Workshop — Spotlight paper.
  • Completed Deep Learning Optimisé sur Jean Zay (IDRIS) — HPC multi-GPU / multi-node training, 4-D parallelism, mixed precision.
  • Talk at EUSIPCO 2025; poster at Hi! Paris Summer School 2025.
  • First Step French Champion of Streetlifting 2024 (−80 kg); former competitive cyclist — Department & Regional Team Champion.