This Week’s Discoveries | 22 January 2019
- Jan-Willem Buurlage
- Sofia Fernandes Gomes
- 22 januari 2019
Niels Bohrweg 2
2333 CA Leiden
- De Sitterzaal
Data partitioning strategies for 3D tomographic imaging
Jan-Willem Buurlage (CWI and MI) is a PhD student in the computational imaging group of Joost Batenburg at CWI, where he works on various topics in real-time 3D tomography. His research focuses on parallel algorithms and distributed-memory methods in inverse problems.
3D imaging is an important tool for industry, medicine, science and culture. Computed Tomography (CT) is a non-destructive technique for imaging the interior of an object. In image reconstruction for CT, a key challenge is dealing with 3D volumes in a computationally efficient way. 3D volumes in tomography can consist of up to 4000 x 4000 x 4000 elements. This means that the data involved just for storing the 3D volume can already be 256 GB in size. In order to achieve reasonable runtimes for CT algorithms, multiple processing elements such as GPUs have to be used. However, depending on how data is assigned to the different processing elements, communication and coordination can quickly become a bottleneck for realizing fast reconstructions. Generally, such data assignment problems are modeled as partitioning problems on (hyper)graphs. In this talk, a geometric partitioning method is presented instead. This partitioning method aims to minimize the required communication between processor elements for 3D iterative reconstruction algorithms.
Cheating belowground interactions
Sofia Fernandes Gomes (CML) studied in University of Lisbon in Portugal, where she graduated in Molecular Biology, and did a MSc in Applied Microbiology. She finished her PhD in October 2018 at the Institute of Environmental Sciences and this talk is a summary of her PhD research. Sofia is interested in the ecological and evolutionary aspects that shape plant and fungal communities.
Mycoheterotrophy is a particular mode of life in which plants obtain carbohydrates from their associated fungal partners, instead of using photosynthesis. The majority of mycoheterotrophic plants exploit mycorrhizal fungi and represent clear examples of shifts from mutualistic mycorrhizal interactions to cheating. Due to the complexity of mycorrhizal interactions and challenges in assessing the outcomes of this symbiosis – both for plants and fungi – mycorrhizal cheating has remained a poorly researched topic in plant ecology. I study the diversity, ecology and distribution of mycoheterotrophic interactions, at species level, population level, and local and global scales. My results highlight the scale-dependent factors that explain the occurrence of mycoheterotrophy. Primarily, mycoheterotrophic plants require at least the presence of their associated fungal partners to persist, which should occur predominantly within humid forests. When these conditions are fulfilled, the balance between soil nutrients, instead of solely absolute concentrations of nutrients, influence the conditions that favour the occurrence of these plants. Lastly, resource competition may contribute to shape the specificity of fungal interactions of mycoheterotrophic plants, which can vary in degree of specialization respectively to each species.