Experience

Assistant Manager Data Analytics & Modeling

PwC Digital Services – PwC Switzerland
July 2017 – now (Zurich, Switzerland)

Responsibilities

All about data (and images), innovation and more! ML, Image processing, Data Analytics, NLP.

Innovation Scientist

AXA Technology Services – Advanced Engineering Lab
July 2016 – June 2017 (Lausanne, Switzerland)

Responsibilities

In this work, I mostly worked on insurance related projects such as claim processing (e.g, water leakage app) and risk assessment projects (e.g. flood detection) using image processing and deep learning techniques as well as AXA image processing toolbox package.

I also used NLP techniques (natural language processing) to predict the technicality of web articles.

Algorithm Developer – Machine learning (freelancer)

Freelance work
August  2015 – June 2016 

Responsibilities

I worked on some image processing and ML projects in IT and financial industry.

Image processing researcher

Instituto Superior Técnico
September  2014 – June 2015 (Lisbon – Portugal)

Responsibilities

I worked on 3D shape reconstruction of face from a single image by using generic face model.

This method used the input image as a guide to shape the generic (reference) model to reach a desired reconstruction.  (using shape from shading + global similarity of faces by considering a single reference model)

Image processing researcher

Instituto Superior Técnico
February 2013 – July 2014 (Lisbon – Portugal)

Responsibilities

  • In this work, I mainly focused on solving the jigsaw puzzle problem through convex optimization. We proposed a new general methodology by considering a rotated version of a piece in a given position of the puzzle as structural element. This method is able to frame the problem on the convex setting with few assumptions, the dimension of the puzzle is the unique requirement. On the other hand, we introduced an approach to caste the general jigsaw problem into a linear programming and find the minimum cost path. (Application of the jigsaw puzzle problem: organize technical drawings in 3-D space; assemble Portuguese tile panels from individual tiles.)
  • Machine learning, image processing and optimization techniques are extensively used in my research.

Modeling the jigsaw puzzle problem as a graph:

(a) Node of the graph is a rotated version of a piece in a given location and edges (blue straight lines) are links between nodes and their neighbours' locations. (b) Possibility of connecting a fix piece to a rotation of another piece. A feasible solution must have an edge per border. (c) A feasible solution: each node has four neighbours at most (one per border), there is one node per location and each piece is selected once.

(a) Node of the graph is a rotated version of a piece in a given location and edges (blue straight lines) are links between nodes and their neighbours’ locations. (b) Possibility of connecting a fix piece to a rotation of another piece. A feasible solution must have an edge per border. (c) A feasible solution: each node has four neighbours at most (one per border), there is one node per location and each piece is selected once.

Scramble images (a)-(c) and corresponding computed solutions (d)-(f).

Visualizing the square jigsaw problem (6×9) :

Freelancer (image processing)

Freelance work
December 2012 – January 2013 (Lisbon – Portugal)
Responsibilities
  • In eye-gaze control in gaming project for tracking eyes using webcam, I used Haar Feature-based Cascade Classifier in OpenCV to detect the eyes.

Machine Learning Scientist

Instituto de Telecomunicações 
September 2011 – November 2012 (Lisbon – Portugal)
Responsibilities
  • In Alzheimer’s disease: Recognition and Image Analysis project, I had a chance to work with large ADNI database. In this work a computer aided diagnosis tool was presented for the study of Alzheimer’s disease based on PET images. The proposed tool applied Principal Component analysis (PCA) to the voxels contained in small regions of interest selected by a physician, in order to reduce dimensionality. Subsequently, classification was performed using Support Vector Machines (using LibLinear & LibSVM libraries).
  • To predict MCI to AD conversion for long-term applied on the ADNI database, I also built a model on AD-Control data using decision boundary and Clinical Dementia Rating (CDR).

☛ Alzheimer’s Diagnosis using PCA and Support Vector Machines in VOIs applied on the ADNI Database

Defined region for different slices of the brain

Defined region for different slices of the brain

Fellowship researcher

Instituto Superior Técnico
December 2009 – May 2010 (Lisbon – Portugal)
Responsibilities
  • I was working on “Design a respirometry system for monitoring wastewater with PLC (PL7) and Labview” project.
  • In this project automated respirometry system was developed from scratch in a laboratory environment, with the aim of obtaining real time data reads and to draw respirometric curves for different types of residual water and biomass from different origins. The system was essentially composed of an airing and mixing tank, where the sample was introduced; and a separate measurement cell where the dissolved oxygen readings took place. The system’s actuators were connected to a software managed controller (programmable logic controller – PLC), and implemented in LabVIEW environment.
  • My responsibilities in this project were to implement of these strategies in LabVIEW environment and resorting to a Schneider Electric PLC.
Overview of the respirometric set-up.

Overview of the respirometric set-up.