Search Results
Climate change and specialty coffee potential in Ethiopia
Current climate change impact studies on coffee have not considered impact on coffee typicities that depend on local microclimatic, topographic and soil characteristics. Thus, this study aims to provide a quantitative risk assessment of the impact of climate change on suitability of five premium specialty coffees in Ethiopia. We implement an ensemble model of three machine learning algorithms to predict current and future (2030s, 2050s, 2070s, and 2090s) suitability for each specialty coffee under four Shared Socio-economic Pathways (SSPs). Results show that the importance of variables determining coffee suitability in the combined model is different from those for specialty coffees despite the climatic factors remaining more important in determining suitability than topographic and soil variables. Our model predicts that 27% of the country is generally suitable for coffee, and of this area, only up to 30% is suitable for specialty coffees. The impact modelling showed that the combined model projects a net gain in coffee production suitability under climate change in general but losses in five out of the six modelled specialty coffee growing areas. We conclude that depending on drivers of suitability and projected impacts, climate change will significantly affect the Ethiopian speciality coffee sector and area-specific adaptation measures are required to build resilience.
Evolutionary design of explainable algorithms for biomedical image segmentation
An unresolved issue in contemporary biomedicine is the overwhelming number and diversity of complex images that require annotation, analysis and interpretation. Recent advances in Deep Learning have revolutionized the field of computer vision, creating algorithms that compete with human experts in image segmentation tasks. However, these frameworks require large human-annotated datasets for training and the resulting “black box” models are difficult to interpret. In this study, we introduce Kartezio, a modular Cartesian Genetic Programming-based computational strategy that generates fully transparent and easily interpretable image processing pipelines by iteratively assembling and parameterizing computer vision functions. The pipelines thus generated exhibit comparable precision to state-of-the-art Deep Learning approaches on instance segmentation tasks, while requiring drastically smaller training datasets. This Few-Shot Learning method confers tremendous flexibility, speed, and functionality to this approach. We then deploy Kartezio to solve a series of semantic and instance segmentation problems, and demonstrate its utility across diverse images ranging from multiplexed tissue histopathology images to high resolution microscopy images. While the flexibility, robustness and practical utility of Kartezio make this fully explicable evolutionary designer a potential game-changer in the field of biomedical image processing, Kartezio remains complementary and potentially auxiliary to mainstream Deep Learning approaches.
Computational design and optimization of electro-physiological sensors
Electro-physiological sensing devices are becoming increasingly common in diverse applications. However, designing such sensors in compact form factors and for high-quality signal acquisition is a challenging task even for experts, is typically done using heuristics, and requires extensive training. Our work proposes a computational approach for designing multi-modal electro-physiological sensors. By employing an optimization-based approach alongside an integrated predictive model for multiple modalities, compact sensors can be created which offer an optimal trade-off between high signal quality and small device size. The task is assisted by a graphical tool that allows to easily specify design preferences and to visually analyze the generated designs in real-time, enabling designer-in-the-loop optimization. Experimental results show high quantitative agreement between the prediction of the optimizer and experimentally collected physiological data. They demonstrate that generated designs can achieve an optimal balance between the size of the sensor and its signal acquisition capability, outperforming expert generated solutions.