EAISI Café | Women in AI

Date
Friday February 11, 2022 from 3:00 PM to 4:45 PM
Location
Online

EAISI CAFÉ | WOMEN IN SCIENCE EDITION

Anyone interested in or working with AI is most welcome to join this online event on International Day of Women in Science, hosted by Patricia Jaspers, Managing Director at EAISI. Researchers will give short presentations on multiple AI topics. The program offers plenty of time for questions and discussion.

Save the date for an inspiring afternoon! 

After registration you will receive a log-in link for MS teams by email.

Program

START

SPEAKER

TITLE

3:00

Patricia Jaspers

Host - Managing Director at EAISI

Opening

KEYNOTE

3:05

Hala Elrofai

Senior Researcher - Mobile Perception Lab, Electrical Engineering Department, AI Program Manager Robotics at EAISI 

Hybrid AI - Trustworthiness and learning beyond training datasets
 

PITCHES

3:25

Iris Kolenbrander
PhD at Biomedical Image Analysis group, Department Biomedical Engineering - TU/e

A mixed-scale dense convolutional neural network for unsupervised joint rigid and deformable image registration

3:35 Sanne Schoenmakers
Assistant Professor at Human Technology Interaction, Department Industrial Engineering & Innovation Sciences - TU/e
Computers, people and the interaction between them
3:45 Leyla Biabani
PhD at Algorithms, Department Mathematics & Computer Science - TU/e
Algorithms for k-Center Clustering with Outliers on Massive Data
SHORT COFFEE BREAK
4:00 Pascalle Wijntjes
PhD at Cardiovascular Biomechanics, Department Biomedical Engineering - TU/e
Hybrid modeling-based prediction of pregnancy complications
4:10 Regina Luttge
Associate Professor at Microsystems, Department Mechanical Engineering - TU/e
Boosting AI’s IQ
4:20 Ekaterina Petrova
Assistant Professor at Information Systems in the Built Environment, Department of the Built Environment - TU/e
A modular AI approach to Digital Twinning in the built environment

WRAP-UP & DISCUSSION

 

Abstracts

HALA ELROFAI

KEYNOTE: Hybrid AI - Trustworthiness and learning beyond training datasets

Perception for highly automated systems (e.g. autonomous robots and vehicles) currently relies on Deep Learning, which lacks robustness, generalisability, transparency, and efficiency. Hybrid AI solutions, combination of Deep Learning and knowledge-driven processes, have the potential to be more interpretable, accurate, robust and data efficient than the ones based on state-of-the-art Deep Learning.  This talk explains the concept of Hybrid AI and discuss case study and some of the submitted proposals.

Iris Kolenbrander

A mixed-scale dense convolutional neural network for unsupervised joint rigid and deformable image registration

Deep learning-based image registration methods struggle with predicting large deformations and require rigidly aligned input images. To tackle this, we propose to combine an encoder and a Mixed-Scale Dense network (MS-D-Net) for joint rigid and deformable image registration. MS-D-Net can capture information at various scales efficiently and is lightweight. Therefore, MS-D-Net is expected to improve registration of large deformations.. We evaluate on pelvic T2-weighted MRIs and compare our method to Elastix and VoxelMorph.

Sanne Schoenmakers

Computers, people and the interaction between them

This talk will be a quick showcase of my currently running projects, passed projects and starting projects. Topics include: deepfakes affect reputation, Brain computer interface for ataxia, self-help for dealing with grief/loss with online app, investigating daily-life issue for adhd to support mental health, intelligent robotic arm design for integration of senses, creative AI to enforce creativity in humans, intelligence calibration for artificial intelligence.

Leyla Bianabi

Algorithms for k-Center Clustering with Outliers on Massive Data

As a concrete problem, we study k-center clustering, where we take outliers into account as these are often present in real-world data sets. Our goal is to design and analyze clustering algorithms on massive data sets in known models such as Massively Parallel Computation (MPC) and Streaming.

Pascalle Wijntjes

Hybrid modeling-based prediction of pregnancy complications

Timely intervention in pregnancy complications leads to a reduction in morbidity and mortality. Therefore, pregnant women with an increased risk must be detected before the 16th week of gestation and start with preventive aspirin use.
Currently existing screening methods perform poorly without extra blood tests. Would adding the physiology of the pregnancy to AI prediction models improve the performance?

Regina Luttge

Boosting AI’s IQ

This pitch highlights the idea behind BayesBrain: an EAISI Exploratory Multidisciplinary AI Research project soon to start. We combine an in-silico Bayesian control agent (BCA) with neural tissue hosted by a microfluidic Brain-on-Chip (BoC). Together, these two very different types of computational units form a hybrid learning system potentially offering novel deep tech solutions. Once interfaced, BCA and BoC are believed to work in unity governed by the Free Energy Principle. Toward this paradigm, can such hybrid system truly boost AI’s IQ?

Ekatreina Petrova

A modular AI approach to Digital Twinning in the built environment

State-of-the-art research in the built environment targets bi-directional Digital Twins capable of predictive maintenance and actuation. Closing the feedback loop between Digital Twins and real-world building systems can be realized through decentralized real-time data processing and Artificial Intelligence (AI). The presented modular approach reconciles statistical and symbolic AI in a decision support system, where unsupervised learning (Association Rule Mining) is boosted by knowledge representation techniques to leverage heterogeneous building data and improve indoor environmental quality, energy efficiency and sustainability in the built environment.

 

Organizer

Eindhoven Artificial Intelligence Systems Institute

EAISI facilitates, stimulates and shares knowledge about AI