Projects
Unraveling neural coding in the visual system by combining human and non-human primate Neurophysiology and Deep Convolutional Neural Networks. KU Leuven
The complexity of our visual environment is represented in the brain through hierarchically organized patterns of activity across the ventral visual cortex. Studies focusing on how exactly information is represented by these neurons and on the specific features being encoded, have been constrained by the choice of stimuli to probe neural activity. Selection of images to discover neuronal feature preferences has traditionally been guided by a ...
Integrating causal Bayesian networks and neural networks for probabilistic reasoning with complex data Ghent University
Probabilistic graphical models, and especially causal Bayesian networks, allow combining data and causal knowledge in a principled way. Based on this framework, we can reason probabilistically about complex systems, even when faced with a significant degree of uncertainty. Despite these benefits, the practical adoption of these models remains limited, due to inability to process data with challenging, but common data properties such as (1) a ...
Distributed Deep Neural Networks for Wireless Neuro-Sensor Networks KU Leuven
Electroencephalography (EEG) is a widely used, noninvasive way to measure the electrical activity of the brain. These signals can be harnessed for various purposes, including the monitoring and analysis of sleeping patterns, epileptic seizure detection and brain-computer interfaces. Traditional EEG requires patients to wear a bulky EEG cap with many wires that are connected to the acquisition device. This means that monitoring the patient’s ...
Decoding speech from the brain using deep neural networks KU Leuven
A growing number of hearing-impaired people benefit from a hearing
aid. Due to the current labour-intensive behavioural diagnostics of
the auditory system, hearing aids are not sufficiently adapted to
individual users, as only a limited number of tests can be conducted
per patient.
To address this, we will develop a new measure of brain activity that
will allow automatic and fine-grained ...
Neural Networks under Epistemic Uncertainty for Robust Prediction KU Leuven
Although artificial intelligence (AI) has improved remarkably over the last few years, its inability to deal with fundamental uncertainty severely limits its application. This thesis will reimagine AI to properly treat the uncertainty stemming from our forcibly partial knowledge of the world. As currently practised, AI cannot confidently make predictions robust enough to stand the test of data generated by processes different (even by tiny ...
Neural Networks as Metamodel for Hygrothermal Simulations of Building Components – Reducing the Calculation Time of Probabilistic Assessments KU Leuven
Simulating the hygrothermal response of a building component often involves many uncertainties, such as the exterior and interior climate, or even the exact geometry and material properties. A deterministic assessment often does not suffice to come to a reliable design decision or conclusion, whereas a probabilistic evaluation includes these uncertainties, and thus allows assessing the hygrothermal behaviour and the related damage risks more ...
Artificial neural networks for sequential thinking Ghent University
Despite impressive pattern recognition successes, today's deep neural networks fundamentally fall short compared to biological brains: they do not possess the ability to engage in explicit sequences of understandable logical thoughts. That is a key property of symbolic AI models, but these in turn lack the strong pattern recognition capabilities of neural networks. The neural networks of the future need to possess both properties. Ultimately, ...
Spatially Adaptive Neural Networks for Computer Vision KU Leuven
Over the past decade, computer vision has seen remarkable progress due to the emergence of data-driven deep learning approaches. Convolutional neural networks (CNN) extract relevant features in an automated way by training on annotated data. As research advances, more complex architectures have more trainable parameters and require more computations. However, executing these models requires powerful hardware, which limits their applicability ...
Deep neural networks as a model of speech perception KU Leuven
It has been shown that when speech signals are presented to a person, they can be decoded from the electroencephalogram (EEG) using linear regression. Unfortunately due to the complex and nonlinear nature of the brain, the correlation between the actual and decoded signal are low and highly variable. In this project, we aim to improve this by leveraging deep learning architectures for automatic speech recognition. Firstly (1), we will build ...