The challenge in embedded machine learning is that the mapping from a DNN to a platform is highly complex and poorly understood. There are too many ways to optimize, simplify, transform and retrain a given DNN, there are too many ways to configure, customize, partition and use a given platform, and we lack understanding how the transformations of DNNs and platforms, in combination, affect the final quality of implementation. The research in this CD Lab will improve our understanding of this design space and reveal yet hidden structures and relations between DNN transformations and platform configuration choices.


The Christian Doppler Laboratory for Embedded Machine Learning addresses the fundamental research questions arising when deep machine learning is deployed in embedded systems with tight resource constraints. The focus of research is on

  1. suitable platforms,
  2. optimization and design flows,
  3. estimation,
  4. matching algorithms and implementations, and
  5. life-long, embedded learning.


The Christian Doppler Laboratory for Embedded Machine Learning takes a leading role in the research on embedded machine learning and becomes one of the world leading research labs for this topic. Key performance indicators are accuracy of the target function (categorization, prediction, etc.), cost, performance and energy efficiency of a performed function, with the objective to develop world leading architectures and methods with

  1. the highest accuracy within an energy budget,
  2. the lowest energy consumption for a given target accuracy and
  3. the ability to do life-long learning. Our scope extends across platforms and specific DNNs. While specific vendors such as Xilinx, Nvidia and Intel will continue to provide strong solutions for their respective platforms, we will leverage their tools and frameworks to identify and implement the best platform-DNN combination for a given application problem.