Close

Aziz Ben Ammar

Software Engineer

Download CV in French

About Me

I was able to benefit from a first professional experience with Inria-Saclay in collaboration with the French Navy on Lidar signal and image processing.
Graduated from a Research Master in Mathematics - specialized in stochastic analysis and applications to finance and an M1 in Quantitative Methods - specialized in mathematical economics and econometrics.
Directly accepted in the second year in Artificial Intelligence and Data System Master at Paris Dauphine University and I finished my fourth optional internship oriented to 3D Medical Image Processing at QuantaCell Lab.
I have also followed intensive training in collaboration with the University of Linnaeus, the University of Paris 13 and the University of Strasbourg.

Experience

IFPEN   

Software Engineer (NLP, CV, Kubernetes, Docker, Django)


  • Intervene on the computational aspects of research projects using data science and learning techniques throughout the project life cycle, from the development of the software chains used to the deployment of an industrial application through the development of prototypes or mvp.
  • Software development and engineering, including Python, webservices and web-applications (containerization, orchestration, Linux systems)
  • ML and DL in the fields of image processing, natural Ianguage processing or other.
  • Product development from prototype to ops.

Inria-Saclay (National Navy)   

Research Engineer

Lidar Signal/Image Processing: BatyNet, DANAE++, Kalman Filter:


  • Implementing several predective machine learning models and choose the 2 best ones:
    Random Forest, MLP, XGBoost (with cross validation methods : LOOCV[Leave one out cross-validation], K-Fold, Stratified cross-validation)
  • Modeling and comparative analysis on standard spatial regularization filters:
    Mean filter, median filter, gaussian filter, bilateral filter, bilateral gradient filter.
  • Modeling and comparative analysis on standard signal denoising filters:
    LowPass filter, HighPass filter, BandPass filter, BandStop filter, Kalman filter, Extended Kalman Filter, Wiener Filter, Hilbert Filter.
  • Implementation of a supervised model and processing of new data using the knowledge available in the vast LIDAR data resources available at Shom.
  • Implementation of an unsupervised model and processing of data from coasts whose characteristics would differ significantly from those available in the previous available in the previous annotated data and where the data available at Shom and already annotated could not be used.
  • Creation of new databases with a certain degree of "sparcity".
  • Training/Validation process through Transfer Learning and state of the art U-Net models in the Bathymetric domain (on Inria's Cluster-GPU).
  • Implementation from scratch state of the art articles without code source on GitHub: BathyNet, DANAE++, etc.
  • Hyperparameter optimization and prediction generation.
  • Presentation of results through statistical methods (curves, input/output difference, confidence interval, etc.)
  • Provide the Naval Hydrographic and Oceanographic Service with a proof of concept in order to introduce modifications/optimizations to the implemented model.

QUANTACELL   

Internship in 3D Image Processing

Instance Segment on Medical Images: GANS, MasK-RCNN, U-Net.. :


  • Creation of new Datasets throw GANS models.
  • Implementation of a Labeling and Deconvoluting Model for new Unlabeled Datasets through a pre-trained semi-supervised model (Naive Baise, MLP, Decision Tree, SVM) on different color spaces (RGB, YUV, etc.) with introduction of white normalization.
  • Creation of masks from XML/NPY files by performing several operations (zoom, dezoom, division of the masks on 4,16,32 tiles, creation of masks by nucleus / cell / image, etc.)
  • Application of Instance Segmentation through several Deep Learning models (U-Net, MaskRCNN, StarDist, Retina-Net, etc.)
  • Implementation of visual functions to detect False Positives / False Negatives / True Positives / False Positives.
  • Integration of the Models after Optimization (Hyperparameters, Data Augmentation, Addition of new varied DataSets in terms of dimension / colors / etc.
  • Deployment of Microservice with all options under the same interactive executable file.

PIXOP    

ML-Ops Internship in Video Processing

Optimization and AWS Deployment of Video Denoising and Resolution Augmentation Models on Python/C++: Autoencoder, ViDeNN:


  • Analyze the image formation process and understanding of the many different mathematical characteristics of noise introduced at various stages.
  • Optimize AutoEncoders and Update state-of-the-art best practices for denoising and Up-Scaling Resolution.
  • Design and implement several different no-hands on machine-learning models in PyTorch to denoising (YUV) color space video.
  • Compare various performance characteristics (training complexity, visual quality, runtime speed, etc.) of the implemented algorithms against each other.
  • Test the best algorithm(s) by processing real-world footage and conduct tests on human test subjects (in collaboration with Pixop) to compare the MOS against Neat Video.
  • Apply Hyper-parameter optimization: Grid-search, Random-search, Gauss-Bayesian-search.
  • Convert mid-level Tf.keras2.x to PyTorch in order to have more relevent plots.
  • Applying Transfer Learning and performance gain analysis.
  • Deployment of the model on the EC2 AWS-bucket and Arduino UNO.

ROAMSMART   

Predictive Modeling and Dynamic Optimization Internship

Development and Deployment of a Micro-service Oriented to Minimization of Phone Operator Expenses Through ML on Python, AWS: RNN (LSTM), M.O.Optimization, M.LReg, Time-Series. :


  • Dynamic optimization of the teleoperators' portfolio (M.O.Optimization).
  • Data analysis and implementation of regression models (M.LReg,Time-Series).
  • Data analysis and implementation of forecasting models (ANN, RNN, LSTM).
  • Simulations and Interpretations of model performance.
  • Deployment of the model on the S3 AWS-bucket.

Central Bank of Tunisia   

Internship in Reinforcement Learning and Dynamic Optimization

Development of a Predictive Bank/Sovereign Default Model on Matlab/R/C++:
AR, ARIMA, SARIMAX, DSGE, VFI, EDM, 2EDM.


  • Modeling of Banking and Sovereign Default in Tunisia (AR, ARIMA, SARIMAX).
  • Data analysis and implementation of forecasting models (DSGE, VFI, EDM, 2EDM).
  • Interpretation of model performance.

Education

The objective of this Master of Mathematics and Computer Science also called Big-Data (IASD) is to offer a solid knowledge in Applied Mathematics as well as in Artificial Intelligence in order to cover all the problems of data processing and analysis. massive that can be found in business.
This training takes place over two years during which two major themes are taught:


  • Data processing, via architectures, algorithms and languages oriented towards large-scale data management.
  • Information processing via learning techniques (Deep Learning and Reinforcement learning) and Artificial Intelligence.

The Quantitative Economics mention aims to train economists at the best level, open and responsive to the plurality of questions and challenges facing the contemporary economy, while having mastery of quantitative tools to address them in terms of data given in particular the developments in the world of decision-makers, both public and private, brought about by the Big Data .
This Master also allows me to:


  • Train in research in economics and reflect on the methods of decision-making by public and private actors.
  • Be open to innovative methods of processing large databases and to advanced computer programming of complex economic problems.
  • Respond to economic issues that arise in the different areas of the economy: Health, public policies, macroeconomics, finance, energy, environment, development, ...
  • Implement statistical and econometric tools in order to obtain reliable and robust answers, making it possible to inform the choices of public or semi-public institutions or private companies.

The goal of this Master is to present the usual continuous-time stochastic processes and to allow students to deepen in directions such as: finance, reliability, Monte-Carlo methods, links between probabilities and partial differential equations :


  • Brownian motion: construction, properties of trajectories
  • Continuous-time martingales, stop theorem.
  • Stochastic integral, Itô's formula. Application to finance (Black-Scholes model).
  • Stochastic differential equations with Lipschitzian coefficients. Links with partial differential equations.
  • Application to options. Variational inequalities and American options.

The Mathematics license allows me to acquire in-depth disciplinary skills in Mathematics (analysis, algebra, probability, etc.), knowing how to organize mathematical reasoning and write rigorously. The development of relational and organizational skills are also at the heart of the training.

El Menzah 6 High School   

June 2008 - June 2009

Baccalaureate Mathematics Section

Intensive Academic Training In collaboration with FST

Numerical Probability

Universities of Linnaeus    and Paris 13  : Simulation and Monte Carlo Methods.

Actuarial

University of Strasbourg   : Risk Theory and Non-life Insurance.

Personal projects

Platypus

Platypus is a framework for evolutionary computing in Python with a focus on multiobjective evolutionary algorithms (MOEAs). It differs from existing optimization libraries, including PyGMO, Inspyred, DEAP, and Scipy, by providing optimization algorithms and analysis tools for multiobjective optimization. It currently supports NSGA-II, NSGA-III, MOEA/D, IBEA, Epsilon-MOEA, SPEA2, GDE3, OMOPSO, SMPSO, and Epsilon-NSGA-II.

View Project

Reinfocement Learning

Master's Practical work project.

View Project

Data Viz--Criteo

Data Viz - M2 IASD - Paris Dauphine University.

View Project

K means Spark

Master's Practical work project in Spark [without using ML.Lib]
Implementation: On Dauphine Private Cluster.

View Project

Data Wrangling

Two Master's Practical work projects for Data Wrangling in Python.

View Project

Recommender System

Filtrage Collaboratif.

View Project

Computer Vision Mask Detection

Project: The project related to the current situation: .Link. It's a competition to detect the presence or not of a mask on an image.

View Project

ML on Big Data Project (scala)

Fast SGD with Adagrad and Momentum, comparisons with basic implementations (batch, SGD, minibatch) (RDD, DataFrames, DataSet).

View Project

Anomaly Detection (scala)

Kafka Flink Click-Display/Impression Anomaly Detection

View Project

Technical skills

Languages


AI / ML / Data Science


General skills

Get in Touch