Initial situation:

The rapid development in the field of machine learning has enabled the applications of deep learning (DL) technologies in various industries such as machine vision. The reproducibility of the training processes of these DL models is crucial to accelerate their use in practice. Containerization technologies such as Docker offer a promising solution to standardize, isolate and make deep learning workflows scalable. By using Docker, developers and data scientists can ensure that their models can be trained under identical conditions, regardless of the underlying infrastructure. The challenge is to develop a training process that is stable, scales with increasing workloads and can be accurately reproduced. This process should run in a Docker container, which means that it can run on any system with Docker, and it should be flexible enough to be used for different deep learning projects.

Tasks:

• Literature research: Identification of challenges and solution approaches in the dockerization of training processes.
• Design of a training workflow: Design of a training workflow for deep learning models that includes all necessary steps from data preparation to the actual training and model evaluation.
• Implementation of the training process as a Docker application
• Scaling: Scaling of the training process, for example through the use of Docker Swarm or Kubernetes
• Comparison with conventional training (reproducibility, duration, ...)

Your Profile:

• Ongoing education in a HTL
• Motivation & reliability

Our offer:

• Assumption of printing costs in case of very good or good success
• Support from a supervisor from the relevant specialist department
• Get to know Fill as a potential employer and contribute your own ideas and knowledge
• Exciting opportunity to supplement your theoretical knowledge with practical experience
• Very good working atmosphere in a multi-award-winning family company

Time period:

Start from now

We look forward to receiving your application!