About

My name is Andrés Hoyos-Idrobo. I’m a Senior Research Scientist at Rakuten Group, Inc. in Paris, where I work in the Search & Relevance team. I design deep-learning-based learning-to-rank systems at scale for Rakuten Ichiba, from custom models to their training pipelines.

Previously, I was a Research Scientist at the Rakuten Institute of Technology in Paris (2017–2025), where I worked on recommender systems (deep click-through-rate prediction, encoder-based models and graph neural networks) and, earlier, on projects with Rakuten businesses: spam detection for Rakuten Viber, and user purchase understanding and customer lifetime value for Rakuten France.

I did my PhD in computer science at the Microsoft Research – Inria Joint Centre (MediLearn project) and Université Paris-Saclay, in the Parietal team, co-supervised by Bertrand Thirion and Gaël Varoquaux. I spent most of those years at NeuroSpin, and defended my thesis, Ensembles of models in fMRI: stable learning in large-scale settings, in 2017.

To get in touch, send me an email. For my full career history, see my LinkedIn profile. ORCID: 0000-0003-1729-1927.

Research interests

I’m broadly interested in machine learning and statistics, and in methods that stay fast and reliable on large, high-dimensional data.

Now: my work is driven by e-commerce problems.

  • Learning to rank and recommendation: deep ranking models at scale, search relevance, click-through-rate prediction, graph neural networks, and counterfactual evaluation from implicit feedback
  • Optimal transport, for re-ranking and for aligning representations
  • Causal inference: bias correction and treatment-effect estimation
  • Machine learning on new compute paradigms, such as annealing machines

Before: during my PhD I worked on statistical learning for brain imaging.

  • Decoding cognitive states from fMRI
  • Fast clustering and randomized methods for dimension reduction
  • Ensembles of models for stable, scalable estimation

Experience

  1. 2026 – present
    Senior Research Scientist
    Rakuten Group, Inc., Paris · Search & Relevance team

    Deep learning to rank at scale for Rakuten Ichiba, from custom models to their training pipelines.

  2. 2021 – 2025
    Research Scientist
    Rakuten Institute of Technology, Paris

    Recommender systems: deep click-through-rate (CTR) prediction, encoder-based models and graph neural networks.

  3. 2017 – 2021
    Research Scientist
    Rakuten Institute of Technology, Paris

    Spam detection for Rakuten Viber; user purchase understanding and customer lifetime value for Rakuten France.

Education

  1. 2014 – 2017
    PhD in computer science
    Microsoft Research – Inria Joint Centre / Université Paris-Saclay · Parietal team, NeuroSpin

    MediLearn project. Thesis: Ensembles of models in fMRI: stable learning in large-scale settings. Advised by Bertrand Thirion and Gaël Varoquaux.

  2. 2012 – 2013
    MVA master's (Mathematics, Vision, Learning)
    ENS Cachan

    Project on radio-frequency pulse design for high-field MRI, advised by Nicolas Boulant.

  3. 2004 – 2010
    B.Sc. in Biomedical Engineering and B.Sc. in Electronic Engineering
    Universidad Autónoma de Occidente (UAO), Colombia

    Advised by Juan Diego Pulgarin.

Software

nilearn

Python library for machine learning on neuroimaging data. I contributed the FReM decoders and ReNA clustering.

FReM · ReNA

Python

ReNA

Reference implementation of Recursive Nearest Agglomeration, a linear-time clustering for structured signals such as images.

Python

spark-dirty-cat

Similarity encoding of dirty categorical variables (strings) for Apache Spark, released by Rakuten.

ScalaApache Spark

Beyond work

I speak Spanish, English and French, and some Portuguese. Outside work, I enjoy drawing, painting and cinema.