Speech Title: Designing High-Performance and Scalable Middleware for HPC, AI and Data Science based on MPI: The MVAPICH2-based Approach

This talk will focus on challenges and opportunities in designing middleware for HPC, AI (Deep/Machine Learning), and Data Science for On-premise HPC and Cloud systems with advances in networking and accelerator technologies. For the HPC domain, we will discuss about the challenges in designing runtime environments for MPI+X programming models by taking into account support for multi-core systems (Xeon, ARM and OpenPower), high-performance networks (InfiniBand and RoCE), GPUs (including GPUDirect RDMA), and emerging BlueField-2 DPUs. Features, sample performance numbers and best practices of using MVAPICH2 libraries (http://mvapich.cse.ohio-state.edu) will be presented. For the Deep/Machine Learning domain, we will focus on MPI-driven solutions (http://hidl.cse.ohio-state.edu) to extract performance and scalability for popular Deep Learning frameworks (TensorFlow and PyTorch), Machine Learning frameworks (cuML) and large out-of-core models. Accelerating Deep Learning applications with Bluefield-2 DPUs will also be presented. MPI-driven solutions to accelerate data science applications like Dask (http://hibd.cse.ohio-state.edu) will be highlighted. Finally, we will outline the challenges and experiences in deploying this middleware to the HPC cloud environments for Azure, AWS, and Oracle.

Speech Title: Trends in AI & Scientific Computing/HPC and Its Convergence

Speech Title: Proteogenomic Analysis of MD2 Pineapple to Improve Genome Annotation

MD2 pineapple (Ananas comosus) is the second most important tropical crop that preserves crassulacean acid metabolism (CAM), which has high water-use efficiency, and is fast becoming the most consumed fresh fruit in the United Kingdom. Despite the significance of environmental efficiency and popularity, until very recently, its genome sequence had not been determined and a high quality annotated proteome has not been available. Here we have undertaken a pilot study analysing the proteome of MD2 pineapple using Liquid Chromatography coupled with Mass Spectrometry (LC-MS/MS) and identified peptides and candidate proteins using SearchGUI and PeptideShaker which combine three different search engines; X!Tandem, OMSSA and Comet. The searches were conducted against two protein sequence databases; one predicted from the official genome sequence annotation, and one predicted from a set of consensus sequences derived from Next Generation Sequencing (NGS) transcriptomics, to help improve the genome annotation. The output from both datasets was further analysed to examine common protein identifications and also those unique to each database. In total, 1,259 proteins in the annotated V3 genome sequenced in 2015 had supporting peptide evidence from the proteomics study. However, 16 cases were found where protein identifications where exclusive to the transcriptome-derived database and were not found in the standard proteome annotated from the V3 pineapple genome sequence. A further 334 transcript-derived protein identifications were found in common with the annotated V3 pineapple genome. The MD2 transcripts were also compared to the MD2 pineapple genome which was subsequently released in 2016 resulting in 260 non overlapping peptides with either of the genomes. In total, these peptide identifications led to 7 high-quality gene candidates that are novel to both of the pineapple genomes but have good gene structure and homologues in monocot plants. The analyses suggest that transcript-derived proteomics is a productive method to discover both novel genes and genetic structure that is currently missing from the V3 pineapple genome annotation and can help improve its annotation.

Speech Title: HPC and Biomimetic (Aerospace Eng.)

High-performance computing has traditionally been the domain of physicists, chemists, and astrophysicists, who have dominated the use of the world's largest computing systems. Biomimetics is a multidisciplinary field that combines principles from physics, computer science, and biological sciences to produce some of the most artificially created biological creatures that are nearly identical to their natural counterparts. HPC systems are now evolving to fulfill biomimetic needs, assisting in the understanding of the complex science behind their motion, particularly at lower size scales, through the development of new hardware and software components that enable more sophisticated data usage. With examples and case studies, this talk focuses on how HPC can help enhance biomimetics research and development.

Speech Title: HPC and Deep Learning

Deep learning is not a new research field; however, it has drawn attention due to breakthroughs that have been achieved recently in many research fields such as Computer Vision (CV) and Natural Language Processing (NLP). Such advancements would not be able to shine without powerful computing capabilities such as Graphical Processing Units (GPUs) which reduce the computational cost and boost up building deep learning models. In this session, we shed some light on leveraging how GPUs utilise parallel computing to speed up the training of Long Short-Term Memory (LSTM), one of the neural network architectures that requires a very long training time and computational power. We are going to demonstrate the NVIDIA CUDA Deep Neural Network library (cuDNN) to speed up the training of an LSTM model and benchmarking the results compared to CPU training.