Pipeline-parallel training has emerged as a popular method to train large Deep Neural Networks (DNNs), as it allows the use of the combined compute power and memory capacity of multiple Graphics... Show morePipeline-parallel training has emerged as a popular method to train large Deep Neural Networks (DNNs), as it allows the use of the combined compute power and memory capacity of multiple Graphics Processing Units (GPUs). However, with the sustaining increase in Deep Learning (DL) model sizes, pipeline parallelism provides only a partial solution to the memory bottleneck in large-scale DNN training. Careful partitioning of the DL model over the available GPUs based on memory usage is required to further alleviate the memory bottleneck and train larger DNNs. mCAP is such a memory-oriented partitioning approach for pipeline parallel systems, but it does not scale to models with many layers and very large hardware setups, as it requires extensive profiling and fails to efficiently navigate the partitioning space to find the most memory-friendly partitioning. In this work, we propose CAPSlog, a scalable memory-centric partitioning approach that can recommend model partitionings for larger and more heterogeneous DL models and for larger hardware setups than existing approaches. CAPSlog introduces a new profiling method and a new, much more scalable algorithm for recommending memory-efficient partitionings. CAPSlog reduces the profiling time by 67% compared to existing approaches, searches the partitioning space for the optimal solution orders of magnitude faster and can train significantly larger models. Show less
This thesis describes the antimicrobial discovery strategy developed in our group, the den Hertog Group at the Hubrecht Institute. It includes a cultivation-based screening approach for novel... Show moreThis thesis describes the antimicrobial discovery strategy developed in our group, the den Hertog Group at the Hubrecht Institute. It includes a cultivation-based screening approach for novel antimicrobial agents from the source of fungi, and a bacterial time-lapse imaging approach for antimicrobial mechanism of action (MoA) identification. With this strategy, we have discovered several interesting antimicrobial agents and have demonstrated the detailed antimicrobial property of two of them, berkchaetoazaphilone B (BAB) and harzianic acid (HA). Show less
Ouyang, X.; Hoeksma, J.; Velden, G. van der; Beenker, W.A.G.; Triest, M.H. van; Burgering B.M.T.; Hertog, J. den 2021
Antimicrobial resistance has become one of the major threats to human health. Therefore, there is a strong need for novel antimicrobials with new mechanisms of action. The kingdom of fungi is an... Show moreAntimicrobial resistance has become one of the major threats to human health. Therefore, there is a strong need for novel antimicrobials with new mechanisms of action. The kingdom of fungi is an excellent source of antimicrobials for this purpose because it encompasses countless fungal species that harbor unusual metabolic pathways. Previously, we have established a library of secondary metabolites from 10,207 strains of fungi. Here, we screened for antimicrobial activity of the library against seven pathogenic bacterial strains and investigated the identity of the active compounds using ethyl acetate extraction, activity-directed purification using HPLC fractionation and chemical analyses. We initially found 280 antimicrobial strains and subsequently identified 17 structurally distinct compounds from 26 strains upon further analysis. All but one of these compounds, berkchaetoazaphilone B (BAB), were known to have antimicrobial activity. Here, we studied the antimicrobial properties of BAB, and found that BAB affected energy metabolism in both prokaryotic and eukaryotic cells. We conclude that fungi are a rich source of chemically diverse secondary metabolites with antimicrobial activity. Show less
