Projects
Continual Learning: how to defy catastrophical forgetting KU Leuven
Lately there have been impressive improvements in the fields of Machine Intellegence and AI. The problem with these recent models is that they are not able to adapt or evolve. This is problematic when the model has to learn a new category or label, a new action or behavior or something else. The only way a static model can adapt and learn something new is by retraining on all previous data and the new data combined. This means that all ...
Monitoring of cutting and processing quality on a forage harvester’ KU Leuven
In the agricultural sector, innovative methods are continuously developed to optimize agricultural processes with the ultimate purpose to obtain, in a sustainable way, a higher (quality of) yield from the same land. Agricultural vehicles should therefore be controlled in an optimal manner: machine settings are often suboptimal. A better processing of crops is realizable if real-time monitoring can be used for analysing the effect of the ...
Bridging the Image and Text Spaces with Neural Network Methods for Multimodal Representation Learning and Spatial Understanding KU Leuven
In the recent years, the emergence of deep learning models has greatly advanced computer vision and natural language processing (NLP). These models allow producing feature representations for images and words. For example, convolutional neural networks (CNN) and Mikolov’s et al. skip-gram model are popular means to extract such features from images and text corpora respectively. These representations are typically employed either in concept ...
Deep Learning Models for Generating Text Summaries from Videos KU Leuven
In recent years, there has been a growing interest in bridging different modalities in the deep learning community. This project aims at developing deep learning models for automatically generating text summaries from videos. Through this project, we expect to find better ways to integrate textual and visual modalities into an end-to-end model. Apart from that, text summarization techniques, including abstractive text summarization and ...
ADOPSYS Vrije Universiteit Brussel
There is a severe shortage in the European photonics industry of graduates and PhDs with broad expertise in modelling and design of industrially relevant optical systems and components. This shortage is aggravated because most applied design methodologies are strongly based on previous experience of the designer. In general, these methods also have a limited efficiency because they heavily ...
Model-based active thermography for contactless quality assessment in the agro-food industry KU Leuven
In the agro-food industry, there is an increasing demand for non-destructive quality inspection tools that allow to inspect every individual product. This is the result of an increased focus on food safety and quality over the past decades. Several fast, non-destructive and contactless technologies have been studied for food quality inspection, many of them based on the interaction of electromagnetic radiation with the food product. ...