The projects are the result of the 2018 ASDI call for proposals. Scheduled to start in 2019, the projects are collaborations with research teams from multiple Dutch academic groups: VU University Amsterdam, Wageningen University, Tilburg University, and University Medical Center Utrecht.
The four projects are the following:
Coastal flooding due to tropical cyclones is one of the world's most threatening hazards with damages up to hundreds of billions of euros per event. The risks of coastal floods are projected to increase in the future due to sea-level rise.
Policy-makers need accurate estimates of current and future flood risks in order to take informed decisions on disaster risk reduction and climate change adaptation. A major scientific challenge is to assess global flood risk with hydrodynamic models that have high resolution and accuracy. This can only be achieved by developing a novel approach that combines cutting-edge disciplinary science and eScience technologies. The aim of this project is developing and validating a computationally efficient, scalable, framework for large-scale flood risk assessment. This framework incorporates two major innovations:
Within the project, they will use the framework to test whether these improvements lead to more accurate estimates of extremes sea levels, inundation extent and flood risk. To this end, they select the North-Atlantic as a case study area. The novel framework is an important step towards improved global assessments of flood risk.
Recent extreme droughts combined with accelerating human exploitation are pushing tropical forests to the point where they cannot recover, making them vulnerable to large unprecedented wildfires. There is as such an urgent need to monitor the recovery capacity of tropical forests. While time series-based break detection approaches have demonstrated potential to measure tropical forest recovery capacity, they have not yet been applied over large amounts of satellite data. The reasons for this are twofold:
This proposal addresses these two critical bottlenecks by:
By adapting and combining these innovative technologies, the researchers intend to measure forest recovery capacity at unprecedented spatial and temporal scales using the new RADAR Sentinel-1 satellite data. By making data and algorithms more accessible and scalable, this project addresses an urgent need of Dutch service providers and will demonstrate the potential of European Sentinel satellite data for large-scale monitoring tropical forests.
Understanding spoken language is an important capability of intelligent systems which interact with people. Example applications which use a speech understanding component include personal assistants, search engines and others. The common way of enabling an application to understand and react to spoken language is to first transcribe speech into text using a speech recognition module, and then to process the text with a separate text understanding module.
The researchers propose an alternative approach inspired by how humans understand speech. Speech will be processed directly by an end-to-end neural network model without first being transcribed into text, avoiding the need for large amounts of transcribed speech needed to train a traditional speech recognition system. The system will instead learn simultaneously from more easily obtained types of data: for example, it will learn to match images to their spoken descriptions, answer questions about images, or match utterances spoken in different languages.
This proposal promises to be less reliant on strong supervision and expensive resources and thus applicable in a wider range of circumstances than traditional systems, especially when large amounts of transcribed speech re not available, for example when dealing with low-resource languages or specialized domains.
Early detection of developmental dyslexia (DD), a specific reading disorder, will enable interventions at an early age, before the onset of formal reading and spelling instruction. Although deviations in early speech/language development have frequently been related to (risk of) DD, none of these markers - vocabulary; auditory brain responses (EEG) - have been successfully used to predict later language/literacy performance at the individual level.
Machine learning (ML) is a technique capable of discovering patterns in data to make such predictions. In the past decade ML has been successfully employed to predict psychosis, disease-course, age, IQ, etc. from measurements such as MRI and EEG. This project will use ML to explore if EEG data on speech processing in infancy can predict the occurrence of later literacy difficulties in individual children.
Application of ML to EEG recordings in infants at risk of DD and low-risk controls requires unconventional, beyond state-of-the-art eScience solutions. An important step of this project is validation of the discriminating patterns in an independent sample from our international collaborators. If successful, this research will open the way to further investigate the problem of prediction of (ab)normal development, ranging from DD, Specific Language Impairment, to autism and psychotic disorders such as schizophrenia.