Deep learning is rapidly becoming a go-to tool for many artificial intelligence problems due to its ability to outperform other approaches and even humans at many problems. Despite its popularity we are still unable to accurately predict the time it will take to train a deep learning network to solve a given problem. This training time can be seen as the product of the training time per epoch and the number of epochs which need to be performed to reach the desired level of accuracy. Some work has been carried out to predict the training time for an epoch -most have been based around the assumption that the training time is linearly related to the number of floating point operations required. However, this relationship is not true and becomes exacerbated in cases where other activities start to dominate the execution time. Such as the time to load data from memory or loss of performance due to non-optimal parallel execution. In this work we propose an alternative approach in which we train a deep learning network to predict the execution time for parts of a deep learning network. Timings for these individual parts can then be combined to provide a prediction for the whole execution time. This has advantages over linear approaches as it can model more complex scenarios. But, also, it has the ability to predict execution times for scenarios unseen in the training data. Therefore, our approach can be used not only to infer the execution time for a batch, or entire epoch, but it can also support making a well-informed choice for the appropriate hardware and model.

Accurate predictive modelling of the growth of microbial communities requires the credible representation of the interactions of biological, chemical and mechanical processes. However, although biological and chemical processes are represented in a number of Individual-based Models (IbMs) the interaction of growth and mechanics is limited. Conversely, there are mechanically sophisticated IbMs with only elementary biology and chemistry. This study focuses on addressing these limitations by developing a flexible IbM that can robustly combine the biological, chemical and physical processes that dictate the emergent properties of a wide range of bacterial communities. This IbM is developed by creating a microbiological adaptation of the open source Large-scale Atomic/Molecular Massively Parallel Simulator (LAMMPS). This innovation should provide the basis for “bottom up” prediction of the emergent behaviour of entire microbial systems. In the model presented here, bacterial growth, division, decay, mechanical contact among bacterial cells, and adhesion between the bacteria and extracellular polymeric substances are incorporated. In addition, fluid-bacteria interaction is implemented to simulate biofilm deformation and erosion. The model predicts that the surface morphology of biofilms becomes smoother with increased nutrient concentration, which agrees well with previous literature. In addition, the results show that increased shear rate results in smoother and more compact biofilms. The model can also predict shear rate dependent biofilm deformation, erosion, streamer formation and breakup.

We present NUFEB (Newcastle University Frontiers in Engineering Biology), a flexible, efficient, and open source software for simulating the 3D dynamics of microbial communities. The tool is based on the Individual-based Modelling (IbM) approach, where microbes are represented as discrete units and their behaviour changes over time due to a variety of processes. This approach allows us to study population behaviours that emerge from the interaction between individuals and their environment. NUFEB is built on top of the classical molecular dynamics simulator LAMMPS (Large-scale Atomic/Molecular Massively Parallel Simulator), which we extended with IbM features. A wide range of biological, physical and chemical processes are implemented to explicitly model microbial systems, with particular emphasis on biofilms. NUFEB is fully parallelised and allows for the simulation of large numbers of microbes (107 individuals and beyond). The parallelisation is based on a domain decomposition scheme that divides the domain into multiple sub-domains which are distributed to different processors. NUFEB also offers a collection of post-processing routines for the visualisation and analysis of simulation output. In this article, we give an overview of NUFEB’s functionalities and implementation details. We provide examples that illustrate the type of microbial systems NUFEB can be used to model and simulate.

No abstract

Urban flood risk modelling is a highly topical example of intensive computational processing. Such processing is increasingly required by a range of organisations including local government, engineering consultancies and the insurance industry to fulfil statutory requirements and provide professional services. As the demands for this type of work become more common, then ownership of high-end computational resources is warranted but if use is more sporadic and with tight deadlines then the use of Cloud computing could provide a cost-effective alternative. However, uptake of the Cloud by such organisations is often thwarted by the perceived technical barriers to entry. In this paper we present an architecture that helps to simplify the process of performing parameter sweep work on an Infrastructure as a Service Cloud. A parameter sweep version of the urban flood modelling, analysis and visualisation software "CityCat" was developed and deployed to estimate spatial and temporal flood risk at a whole city scale -far larger than had previously been possible. Performing this work on the Cloud allowed us access to more computing power than we would have been able to purchase locally for such a short time-frame (∼21 months of processing in a single calendar month). We go further to illustrate the considerations, both functional and non-functional, which need to be addressed if such an endeavour is to be successfully achieved.