Johannes Kiechle

This work showcases a variational graph autoencoder disentanglement using 3D mesh data. For further information, please refer to the corresponding paper which can be found in the doc directory of the repository.

The overall goal of this project was about developing an algorithm using deep learning, which is capable of predicting whether a 3D-model is printable or not. In this context, printable refers to the context of additive manufacturing. All work is done in cooperation with the Data Analytics & AI team of Volkswagen Data:Lab Munich carried out in the context of the TUM Data Innovation Lab.

This work showcases a holistic approach how to perform anomaly detection for rail traffic data of Finland. Therein, a so-called Isolation Forest is trained in order to detect abnormal data samples.

In this project, an asynchronous value iteration while heavily using OpenMPI is presented. This is done for the purpose of distributing the entire workload across several physical machines. Different strategies are implemented and benchmarked accordingly.

This project implements an asynchronous value iteration for a Markov Decisioin Process (MDP) using Bellman’s principle of optimality. Python (for convenience purposes) and C++ (for performance purposes) are linked using CFFI. Furthermore, OpenMP is used in order to parallelize the value iteration process.

This project showcases a producer-consumer model using mutual exclusions (Mutex) in C++ in order to draw the well known Barnsley fern. Therein, the computational workload can distributed among several workers in order to speed up the entire process. The number of workers is selectable and can be adjusted according to the processing hardware at hand.

In the context of a university lecture called “Applied Machine Intelligence” at Technical University Munich (TUM), chair of data processing, in a group of 8 students, we have been investigating the impact of COVID-19 on the climate change. In order to draw inference on collected data, various machine learning approaches were used.

This project was carried out as part of the computer vision lecture at the Technical University of Munich (TUM) in the summer semester 2020. The goal of this project was to implement a person image segmentation using conventional computer vision methods.

The program supports research-oriented stays abroad for master’s students in the fields of artificial intelligence, computer science, and related disciplines. The official certificate can be found here.

The group challenge which I’ve been assigned to was: “Connect Real World Scenarios to the Digital Twin of an Autonomous Vehice in Real Time”. Personal learning outcomes: Collecting as much information as possible done by a smart infrastructure in real-time and providing this additional knowledge to all possible road users will be for sure one of the key components of future autonomous car deployments. The award in PDF file format can be found here.

Outline of modules:

• Introduction to the LRZ Compute Cloud
• Introduction to Container Technology & Application to AI
• Introduction to the LRZ AI Infrastructure
• Intel AI HPC Workshop: Machine Learning Module
• Intel AI HPC Workshop: Deep Learning Module
  Each 3,5h course module was offered as combination of lectures, demos and/or hands-on sessions. The certificate in PDF file format can be found here.