Research Associates (m/f/d) for Embedded Foundation Models & Neuromorphic Edge AI

Environment

Are you interested not only in working with existing AI models but also in exploring how future foundation models can be efficiently run on resource-constrained edge systems? Are you interested in modern deep learning architectures and want to develop new approaches at the intersection of foundation models, neuromorphic AI, and embedded computing?

In the Embedded Systems and Sensors Engineering division at FZI, you’ll work on new AI architectures for industrial edge systems and explore how large pre-trained models can be optimized in terms of computational effort, memory requirements, latency, and energy efficiency. In doing so, you’ll investigate both classical and neuromorphic and hybrid model architectures and further develop them through hardware-software co-design for future AI processors.

Your research will take place within the framework of national and international research projects in collaboration with leading industry and research partners. This position offers you the opportunity to establish your own scientific research profile and pursue a Ph.D. at the Karlsruhe Institute of Technology (KIT). In this environment, a wide range of research questions arise, which you can immediately address as a doctoral candidate under the supervision of Prof. Dr.-Ing. Dr. h. c. Jürgen Becker.

Responsibilities

• You will work on research and development contracts as well as publicly funded research projects.

• You will research and develop resource-efficient Time Series Foundation Models (TSFM) for local execution on embedded and edge systems.

• You will analyze current foundation model architectures and investigate how their model size, memory requirements, computational cost, latency, and energy consumption can be reduced.

• You will develop and evaluate classical, neuromorphic, and hybrid model architectures, such as those based on transformers and spiking neural networks.

• You will investigate techniques such as quantization, pruning, knowledge distillation, and hardware-aware training to tailor models specifically to the characteristics of future edge AI hardware.

• You will collaborate with our researchers in the divisions of processor architecture and hardware development on hardware-software co-design and investigate which model and hardware architectures are particularly efficient for different AI workloads.

• You will implement and evaluate your approaches using modern ML frameworks such as PyTorch and deploy them on embedded and edge platforms as well as newly developed AI hardware.

• You will further develop your research using real-world industrial data and applications and compare your approaches in terms of model quality, generalizability, latency, and energy efficiency.

• You will publish your scientific results at international conferences and in academic journals and use them to develop a doctoral research topic at KIT.

• You will work closely with industrial companies, universities, and other research partners on national and European research projects.

Requirements

• You have earned a master’s degree with high honors in computer science, artificial intelligence, electrical engineering, information technology, technical computer science, computer engineering, or a related degree program.

• You have a solid understanding of machine learning and deep learning and have already gained practical experience with neural networks.

• You are proficient in Python and, ideally, have already worked with deep learning frameworks.

• You are interested in understanding how modern AI models work internally and how their architecture can be optimized for different hardware and resource requirements.

• You enjoy scientific work and would like to pursue a Ph.D. as part of your role.

• You work independently and in a structured manner, contribute your own ideas, and are interested in developing new approaches in collaboration with researchers from various disciplines.

• You can communicate confidently in English. Good German skills are an advantage for collaborating with our industry partners, but are not a mandatory requirement if you have a very strong technical profile.

Ideally, you also have experience in one or more of the following divisions:

• Transformers and Foundation Models

• Time-Series Machine Learning or Time Series Foundation Models

• Model compression, quantization, pruning, or knowledge distillation

• Efficient AI or Edge AI

• Spiking Neural Networks or Neuromorphic Computing

• Hardware-Aware Machine Learning

• Embedded AI and Neural Network Deployment

• C/C++, Embedded Linux, or hardware-close software development

• AI Accelerators 

Application

Have we piqued your interest? Then apply to join our team.
We look forward to getting to know you!

If you have any questions, please email your application materials to Dr.-Ing. Victor Pazmino Betancourt (pazmino∂fzi.de)