High-performance Deep Models for Embedded Devices

Deep networks have shown great performance in solving complex tasks in different application domains. Existing literature demonstrates that the performance of neural networks improves with increasing number of parameters, which usually results in an increased network width measured by the number of neurons in each layer.

In this project we would like to enable efficient implementation of deep networks enjoying such optimization techniques on an embedded device, such as Arduino Nano.

Download as PDF

Student Target Groups:

  • Students in ICE and Computer Science.

Thesis Type:

  • Master Project / Master Thesis

Goal and Tasks:

  • Literature review on deep learning optimizations to improve model accuracy, minimize inference time and optimize memory allocation.
  • Implement a vanilla inference of a pretrained model on Arduino Nano (the platform supports TensorFlow Lite, but you will also do it from scratch).
  • Test different model-specific optimizations.
  • Test different hardware-specific optimization.
  • Summary of the results in a written report, oral presentation.

Recommended Prior Knowledge:

  • A good knowledge of embedded systems.
  • Basic knowledge of deep learning or an interest to learn and understand the basics.
  • Programming skills in Python and C.
  • Your talent (very important!)


  • a.s.a.p.