Adrià Arrufat
Machine Learning Engineer at B*Factory
Seoul, South Korea
Summary
🤩
Rockstar🎓
Top SchoolAdrià Arrufat is a Machine Learning Engineer based in Seoul with 13 years of experience bridging video-compression research and applied ML. He holds a PhD in video compression and brings strong C++, Python, Unix and image-processing skills from research roles at Orange and Technicolor to industry positions at OMNIOUS.AI, Intel and currently B*Factory. An active contributor to core C++ ML tooling and developer tools, he has worked on dlib (including adding a per-channel loss and integrating pretrained ResNet50) and on the Kakoune editor, showing comfort with both low-level libraries and developer ergonomics. Notably, he has applied practical model adaptations—such as replacing dropout with multiplication layers for efficiency—illustrating a focus on squeezing performance from models and code for real-world telecom and imaging workloads.
13 years of coding experience
10 years of employment as a software developer
Doctor of Philosophy (Ph.D.), Doctor of Philosophy (Ph.D.) at INSA RennesMaster's degree, Master's degree at UPC - ETSETB TelecomBCN
Stackoverflow
441reputation
851kreached
5answers
1question
Github Skills (36)
editors10
c-plus10
c-language10
text-editor10
machine-learning10
editor10
zig10
deep-learning10
texteditor10
c-plus-plus10
resnet10
kakoune10
code-editor10
computer-vision10
dlib10
Programming languages (24)
Github contributions (5)
davisking/dlib
May 2019 - Nov 2022
A toolkit for making real world machine learning and data analysis applications in C++
Role in this project:
Backend Developer
Backend DeveloperContributions:178 reviews, 134 commits, 192 PRs in 3 years 7 months
Contributions summary:Adrià's commits primarily focused on addressing pedantic warnings and improving code quality within the dlib library. They made changes to various headers and source files, including `matrix_math_functions.h`, `global_optimization/global_function_search.cpp`, and `dnn/loss.h`, indicating involvement in core library components. The contributions involved fixing warnings, potentially related to code style or optimization issues, and the addition of a new loss function named `loss_mean_squared_per_channel_and_pixel`.
data-analysispythondata-sciencedeep-learningc-plus-plus
davisking/dlib-models
Nov 2019 - Feb 2020
Trained model files for dlib example programs.
Role in this project:ML Engineer
ML EngineerContributions:1 review, 12 commits, 2 PRs in 2 months
Contributions summary:Adrià primarily focused on integrating and modifying pretrained deep learning models within the dlib framework. They replaced dropout layers with multiplication layers for efficiency, and integrated a pretrained ResNet50 model from ImageNet, demonstrating proficiency in model adaptation. Further, they refined the example code to leverage the pretrained weights and fine-tune the model by adjusting learning rates, thereby exhibiting a grasp of deep learning model fine-tuning techniques.
deep-learningface-recognitionmachine-learningprogramsdlib
Find and Hire Top DevelopersWe’ve analyzed the programming source code of over 60 million software developers on GitHub and scored them by 50,000 skills. Sign-up on Prog,AI to search for software developers.