Dr Arnaud Nguembang Fadja



Profession: Head of Machine Learning

Research Department, Paragon semvox GmbH

Paragon semvox GmbH

Konrad-Zuse-Str. 19, D-66459 Limbach (Saarland), Germany

Tel: +49 (0) 6841 80 90 10 Fax: +49 (0) 6841 80 90 10

E-Mail: arnaud.fadja.n@gmail.com

Paragon semvox GmbH

Dr Arnaud Nguembang Fadja

Projects

2023-today
Machine Learning Operations platform for driver behavior adaptivity

Goal: Develop a cloud platform based on the Azure platform. The platform presents an efficient space for automatic training, validation, testing, and monitoring of an ML model that is able to learn user preferences. The platform provides, Data Engineering, Model development, Continuous Development/ Integration, model monitoring

  1. Task: MLOPS platform
  2. Role: Project Lead
  3. Tools management: Jira, git, Bitbucket, Confluence, MLOPS, docker
  4. Libraries: Scikitlean
  5. IDE: Visual Studio Code
  6. Implemented stages: Data acquisition, data process, data versioning (DVC), Automatic training pipeline, ML deployment as service on Azure

2022-today
Machine Learning system embedded in a smart assistant for learning users driving behavior

  1. Task: Build an ML model that is able to learn and adapt to a user-driving behavior
  2. Role: Project Lead
  3. Tools management: Jira, git, Bitbucket, Confluence, MLOPS, docker
  4. Libraries: Scikitlean
  5. IDE: Visual Studio Code, eclipse
  6. Machine learning Models: Skope rule, Random Forest, Decision Tree, Aleph, Neural Networks

2021-2022
Neural-Symbolic Systems for Identifying COVID-19 Patients in a Critical State

The aim of the project is to implement an AI model which predicts if a Covid-19 patient arriving in the hospital will end in a critical state and therefore will need intensive care. The model relies on a
Decision Tree, 3D Convolutional Neural Network, and Hierarchical Probabilistic Logic Programs.

2020-2022
Trip (sudden stop) Prediction in gas turbine using Decision Tree, Random forest and (Recurrent) neural Networks

  • Task: In this project in collaboration with other computer science and mechanical engineers, I was in
  • charge of designing and implementing an intelligent system for predicting a Trip (sudden stop) in a
  • gas turbine. We investigated different machine learning models such as Neural Networks, Decision/
  • Extra Trees, and Random Forest. 
  • Tools management: Trello, git, Bitbucket
  • Role: Scrum Master
  • Libraries: Sklearn, Matplotlib, Numpy, TensorFlow, Keras, Jupiter Notebooks
  • IDE: Visual studio code, TexMaker
  • Programming Language: Python, Latex
  • Machine learning models: Recurrent neural Networks, Decision/Extra Trees, and Random Forest.
  • Publication: paper under review

2019-2021
Identification of natural selection in genomic data with Deep Convolutional Neural Networks

Build a supervised Machine Learning for classifying genomic data that represent portions of genomic sequences of individuals belonging to a certain population. A Convolutional Neural Network is used to determine if the genomic sequences bear traces of a process of natural selection. Training performed on simulated data shows that the proposed model can accurately predict neutral and selection processes on portions of genomes taken from real populations with almost 90% accuracy.

  1.  Tools management: Trello, git, Bitbucket
  2.  Role: Machine Learning Engineer
  3.  Libraries: Tensorflow, Keras, Matplotlib, Numpy, Google Colab
  4.  IDE: Visual studio code, TexMaker
  5.  Programming language: Python, Latex
  6.  Machine learning Models: Convolutional neural network
  7.  Publication: The paper is published in BioData Mining journal, here

2016-2021
Integration of Deep Learning and Probabilistic Logic Programming: Explainable AI

  • Task: Build explainable AI systems by combining Deep Learning (DL) and Probabilistic Logic Programming
  • Toolkit: Cineca, SWI-Prolog
  • Role: Machine Learning Engineer 6 Researcher
  • Programming languages: Python, C, Prolog, Latex, HPC Slurm
  • IDE: Visual studio code, TexMaker
  • Machine learning models: Neural Networks, Arithmetic Circuits, Hierarchical Probabilistic Logic Programming
  • Web application: Phil on SWISH. The manual is available here.
  • Publication: The paper is published in the Machine Learning journal, https://rdcu.be/cmCIA

2020-2022
Morphological Classification of selected medicinal plants leaf using Convolutional Neural Networks

Task: Build a computer vision system embedded in a mobile phone for classifying medicinal plant leaves.
Tools management: Trello, git, Bitbucket+Libraries: Tensorflow, Tensorboard, Keras, Matplotlib, Numpy, Google Colab
Role: Machine Learning Engineer and project manager
IDE: Visual studio code, TexMaker
Machine learning Models: Convolutional neural networks, AlexNet, VGGNet
Publication: paper submitted  and accepted to Biodataming journal https://biodatamining.biomedcentral.com/

2020-2021
Neural networks on FPGA platform

Task: The aim of the project is to implement different neural network architectures on an FPGA
platform based on Xilinx Ultrascale for real-time image processing applications. Performance, both in
terms of execution time and efficiency (accuracy) are compared with those running on a GPU-
based platform. The main objective is to use the FPGA platform as an accelerator for the inference
phase in Deep Neural Networks.
Role: Machine Learning Engineer
Tools management: Trello, git, Bitbucket, google meet
Libraries: Tensorflow, Keras Vitis AI
IDE: Visual studio code
Programming Language: Python
Machine learning models: Deep Neural Networks

2016-2018
Defect identification in pharmaceutical and cosmetic products

  • Task: Build a computer vision system at the end of a product chain control that identifies and
  • classifies defects including particles of plastic/rubber/glass, fibbers; bubbles in pharmaceutical and
  • cosmetic products.
  • Role: Machine LearnignEngineer
  • Libraries: HALCON, Tensorflow
  • IDE: Visual studio code, TexMaker
  • Programming languages: Custom Halcon programming language, C++, Latex
  • Machine Learning models: Multilayer Perceptron, Convolutional Networks
  • Publication: http://ceur-ws.org/Vol-2272/paper1.pdf

2017-2019
Fruits (date & Tomatoes) sorting with Convolutional Neural Networks

Task: Build a computer vision that classifies dates (good and lost skin) and tomatoes fruit
Role: Machine Learning Engineer
Tools management: Trello, git, Bitbucket
Libraries: Tensorflow, Tensorboard, Keras, PyTorch, CNTK
IDE: Visual Studio Code
Machine learning Models: Convolutional neural network, Faster RCNN

2017
Development of research personal website using Ruby On Rails

  • Task: Design and implement a researcher's website
  • Role: Web developer
  • Tools management: Trello, git, Bitbucket
  • Platforms: Heroku, Amazon S3
  • Libraries and Frameworks: VueJs, Quasar, Ruby on Rails
  • IDE: Visual studio code, RubyMine
  • Programming Language: Ruby
  • Final product: http://www.arnaud-fadja.it/