Im Folgenden finden Sie eine Aufstellung der zur Verfügung stehenden Themen. Die angegebene Literatur versteht sich als Startlektüre und weitere Literatur sollte selbstständig recherchiert wertden.
The suggested topics are listed below. The literature indicated here is intended as a starting point only - further literature should be researched independently.
Accelerator technologies, that work alongside CPUs of the compute servers for accelerating certain regions of an application requiring a large amount of numerical operations, are becoming more broadly used and adopted in HPC systems. This topic discusses different contemporary accelerator technologies (GPUs, TPUs, etc.) that are used for boosting the performance of various machine learning applications. In addition, the topic examines the energy-efficiency of the considered accelerator technologies, provides an analysis of state-of-the-art benchmarks used for their evaluation, and suggests a summary outlining the main differences of the considered accelerators.
The topic provides an introduction to FPGAs, outlines their purpose and usability as well as discusses different design flows and concepts.
The topic introduces Arm Scalable Vector Extension version 2 (SVE2) which builds on top of the SVE (v1) - technology utilized by the processors of today's fastest (as of November 2020 Top500 rankings) supercomputer: Fugaku. SVE2 is used by Arm's latest ARMv9-A architecture for further processing enhancement of ML workloads. The topic outlines the fundamentals of SVE2 architecture, describes the new features, as well as discusses the power-efficiency aspects.
The topic discusses the supercomputing today by analyzing current TOP500 list and identifying the major trends. Additionally, it describes Fugaku system, deployed at RIKEN Center for Computational Science (Japan) and currently recognized as the world's fastest supercomputer (TOP500 November'20 rankings). The topic discusses the system design aspects, describes its compute and cooling infrastructures, and examines the energy-efficiency.
The topic discusses different contemporary NVM technologies, introduces their design, architecture, and outlines corresponding applications.
This topic introduces the concept of ensemble learning - a technique for combining the results of several models trained on the same data set for the increase of the overall prediction accuracy.
This topic introduces the concept of transfer learning, discusses its theory and the statistical guarantees.
This topic discusses machine learning paradigms beyond classical supervised and unsupervised learning, and outlines best practices for algorithm design taking advantage of such learning attempts.
Topic discusses the aspects of computational speed and data storage improvements with the help of quantum-enhanced machine learning. This topic also introduces the TensorFlow Quantum library.
Data sanitization is an essential step when building effective machine learning based models. This topic discusses different unsupervised learning methods for outlier detection, analysis their differences as well as outlines their advantages and disadvantages.
This topic continues the discussion on anomaly detection. It investigates the cases where in addition to large set of unlabeled data, one has an access to a small labeled data. The topic discusses the possibilities of maximizing the advantage of the available labeled data using semi-supervised learning approach for the overall increase of the classification accuracy.
The topic introduces a machine learning based transcompiler and discusses its generalization.
This topic discusses recent machine learning based approaches that transform human drawn sketches to images.
The topic introduces the edge computing paradigm, outlines current trends and challenges, and discusses the impact this distributed computing paradigm might have on future of the data centers.
This topic discusses ways for deploying a secure and robust storage cloud using open-source frameworks on Raspberry Pi. The report provides a detailed analysis of the required open-source frameworks as well as evaluates their ease of deployment, security, extendability as well as assesses the efforts required for operation and maintenance.
This topic discusses various strategies and techniques for analysing the performance of large-scale cloud applications.
This topic discusses contemporary container orchestration/management platforms and outlines their advantages and disadvantages. Furthermore, the topic touches upon the aspects of deployment and operation challenges, as well as discusses how those help with the development of AI microservices.
Nowadays, when checking emails, when purchasing from an online store, when reading a document, we always expect to have a time-efficient search engine working for us flawlessly in the background. Moreover, we typically expect these engines to be intelligent enough to support us with various manipulations with our (pending) search results (e.g. sorting, suggesting, etc.). Elasticsearch is an open-source search engine library developed in Java. This topic introduces the basic concepts and principles of Elasticsearch, and outlines how typical search problems can be tackled with the help of Elasticsearch.