Sergio Tascon Morales

I'm a

Download CV

About

I am a highly skilled and ambitious professional specialized in computer vision and NLP for medical data. Known for my structured approach and meticulous attention to detail, I consistently deliver high-quality results within specified timelines. My strong teamwork abilities enable me to thrive in collaborative environments, where I value and integrate diverse perspectives, personalities, and cultural backgrounds to drive innovative solutions.

Education and Experience

Education

Ph.D. Student

2020 - 2024

University of Bern, Switzerland

My research revolved around the topic of Visual Question Answering (VQA). Most of my work was focused on consistency for VQA.
Thesis title: Spatial Awareness and Logic for Robust Visual Question Answering.

Deep Learning
NLP
LLM
Multimodal Learning
Transformers
Dataset Design
Python
PyTorch
German

Erasmus Mundus Joint Master in Medical Imaging and Applications (MAIA)

2018 - 2020

University of Burgundy, France
University of Cassino, Italy
University of Girona, Spain

This international master's program with students from 15 countries took place in France, Italy and Spain.

Machine Learning
Computer Aided Diagnosis
Segmentation
Classification
Registration
Robotics
Signal Processing
Python
Tensorflow
PyTorch
C++
Italian
French
Catalan

Exchange semester

2016

Ilmenau University of Technology, Germany

Exchange semester during the Young Engineers program of the DAAD.

Programming
Signal Processing
MATLAB
Python
German

Electronic Engineering

2011 - 2017

Universidad del Valle, Colombia

A very varied program in which theoretical knowledge was always combined with practical experiments and projects.

Programming
Machine Learning
Computer Vision
Signal Processing
Circuit Design
Telecommunications
Control Systems
C
MATLAB
Java
Python
English
German

Professional Experience

Postdoctoral Researcher

2024 - present

ETH Zurich, Zurich, Switzerland

Research and engineering position in collaboration with Scanvio Medical for the democratization of expert ultrasound especially for the diagnosis of endometriosis from ultrasound data.

Deep Learning
Ultrasound
Segmentation
Python
PyTorch
Bazel

Research Assistant (60%)

2024

University of Bern, Switzerland

Research position for the exploration of further topics in Visual Question Answering. Additional tasks included the supervision of a Master student during the development of his thesis in the topic of VQA for ECG signals.

Deep Learning
NLP
LLM
Multimodal Learning
Transformers
Python
PyTorch

Research Intern

2020

Mediri GmbH, Germany

I developed my master thesis titled “Multiple Sclerosis Lesion Segmentation Using Longitudinal Normalization and Convolutional Recurrent Neural Networks.” This work received the award of best MAIA master thesis.

Segmentation
Longitudinal Data
Deep Learning
Recurrent Neural Networks
Python
PyTorch
German

Research Intern

2016

Endress & Hauser, Germany

Internship under the supervision of two senior developers. My work focused mainly on the improvement of software for linearization of pressure sensors, as well as the evaluation of Python for future use in the company.

Electronic Simulation
Python
MATLAB
German

Skills

Programming Languages

Python
C/C++
Matlab

Libraries / Frameworks

PyTorch
PyTorch Lightning
Tensorflow + Keras
HuggingFace
LangChain
OpenCV
NLTK
NumPy
SpaCy
Scikit Learn
Pandas

Tools / Platforms

Git
Azure
SLURM
VSCode
ITK-Snap
LaTex

Languages

Spanish - Native
English - Advanced
German - Advanced
Swiss German - Intermediate
Italian - Intermediate
French - Basic
Russian - Basic
Turkish - Basic

Publications

Localized Questions in Medical Visual Question Answering

MICCAI 2023
Vancouver, Canada

We introduce a method to answer questions about user-defined regions in an image as opposed to questions about the entire image.

Logical Implications for Visual Question Answering Consistency

CVPR 2023
Vancouver, Canada

A method to improve consistency in VQA by integrating logical relations between pairs of question-answers into the tratining process.

Consistency-Preserving Visual Question Answering in Medical Imaging

MICCAI 2022
Singapore

This approach uses a regularizer to improve consistency in VQA for medical images. The idea is to penalize inconsistent cases during training.

Multiple Sclerosis Lesion Segmentation Using Longitudinal Normalization and Convolutional Recurrent Neural Networks

MLCN Workshop, MICCAI 2020
Lima, Perú (Virtual)

Longitudinal annotations are leveraged to improve segmentations of MS lesions. This is achieved with a combination of UNets and convolutional LSTMs.

Awards

Best MAIA Master Thesis

2020

University of Burgundy, University of Cassino, University of Girona

For the thesis entitled Multiple Sclerosis Lesion Segmentation Using Longitudinal Normalization and Convolutional Recurrent Neural Networks.

Best MAIA Master Student

2020

University of Burgundy, University of Cassino, University of Girona

Award for best student in the third batch of the Master in Medical Imaging and Applications (MAIA).

Erasmus+ scholarship

2018

European Union

Scholarship to study the Master in Medical Imaging and Applications (MAIA) in 3 European Countries.

Bachelor final project with distinction

2017

Universidad del Valle, Colombia

Project entitled Method for the Non-invasive Measurement of the Fetal Electrocardiogram.

Scholarship Young Engineers

2015

German Academic Exchange Service (DAAD)

Scholarship to study one semester at a German university and do an internship at a German company.

Best high school graduate

2009

Institución Educativa Inmaculada Concepción

Award as best graduate.

Hobbies

Violin


Drawing


Reading


Hiking