GPU accelerated computing server for artificial intelligence

我们的目标是获得一个GPU加速的科学计算服务器，由施特拉尔松德应用科学大学应用计算机科学研究所（IACS）的“人工智能和机器学习”能力中心托管和运营。通过专门的人工智能研究小组利用服务器进行研究，开发和应用，旨在加快全面的深度神经网络的训练，分析和操作，从而实现更长时间的训练更大的网络，更多的数据和更广泛的参数扫描参数优化。特别是，该服务器将用于各种研究和应用领域，其目标如下：-模拟生物学上更合理的尖峰神经网络的大型网络，以开发新的学习算法，并有望转移到神经形态平台-支持深度神经网络的分析，以使AI可用于安全关键型应用（而不是通常的黑盒方法）-模拟尖峰神经元的大型网络，以建模和模拟认知过程，并从中学习声学信号处理。对于机器学习在金融、医学、生物技术和工业4.0中的应用，以在更短的时间内使用更大的数据集进行网络训练，并且部分地以真实的时间实现网络应用。- 用于贝叶斯推理和概率编程：为自动推理提供更快的模拟、建模和分析工具-实时分析视频序列，例如，在嘈杂环境中增强唇读阅读的自动语音识别。2.通过应用计算机科学研究所（IACS）内新成立的能力中心“人工智能和机器学习”，对整个机构的团体进行跨部门支持。这里的目的是支持人工智能技术在研究小组中的应用，这些研究小组的核心研究和开发活动并不以人工智能本身为中心。鉴于人工智能技术在工业、研究和开发中的日益普及，这是必要的。3.提高应用科学大学对当地和区域中小企业（SME）的吸引力，使其成为人工智能技术转让项目的首选高级合作伙伴。为此，应用科学大学有必要处理竞争性计算资源，以便不需要访问外部计算资源。

We aim to acquire a GPU-accelerated scientific computation server, hosted and operated by the Competence Centre "Artificial Intelligence and Machine Learning" within the Institute for Applied Computer Science (IACS) at the University of Applied Science Stralsund.The aim with the planned GPU-accelerated computation server is three-fold:1. Utilisation of the server through dedicated AI research groups for research, development and application purposes with the aim to speed up training, analysis and operation of comprehensive deep neural networks, thus enabling longer training of bigger networks with more data and more extensive parameter sweeps for parameter optimisation. In particular the server will be used in various research and application areas with the following aims:- to enable the simulation of large networks of biologically more plausible spiking neural networks for the development of new learning algorithms with the promise of transfer to neuromorphic platforms- to support the analysis of deep neural networks in order to make AI explainable for safety-critical applications (as opposed to the usual black box approach)- to simulate large networks of spiking neurons in order to model and simulate cognitive processes and learn from them for acoustic signal processing.- for applications of machine learning in finance, medicine, biotechnology, and industry 4.0, to enable network training with larger data sets in shorter time, and in part to enable network applications in real time. - for Bayesian inference and probabilistic programming: to provide faster simulation, modeling and analysis tools for automatic inference- to analyse video sequence in real-time, for example to enhance automatic speech recognition from lip reading in noisy environments. 2. Cross-sectional support of groups across the institution through the newly founded competence centre "Artificial Intelligence and Machine Learning" within the Institute of Applied Computer Science (IACS). The intention here is to support the uptake of AI techniques also in research groups whose core research and development activities do not centre about AI per se. This is necessary in the light of increasing pervasiveness of AI techniques in industry, research and development.3. Increasing the attraction of the University of Applied Sciences to local and regional small and medium-sized enterprises (SMEs) as the preferred senior partner in AI technology transfer projects. For this it is necessary that the University of Applied Science dispose over competitive computational resources so that there is no need to access external computational resources.