Bachelor/Master Thesis: Efficient Shallow Neural Networks for Automatic Dietary Monitoring

Background

Neural Networks outperform traditional machine learning approaches in many applications. This is very often caused by high computational efforts of deep models. In wearable contexts, models have to be very efficient if they are run on an embedded system to save battery life. These constraints make the design of neural network models particularly challenging. Recently, gated recurrent units (GRU) have been used in the context of automatic dietary monitoring (ADM) to detect eating events using audio data. The goal of this thesis is to investige GRUs theoretically using data from eating events and to implement an embedded GRU (eGRU) on a microcontroller. Depending on the student’s prior knowledge and interests, the work can be focused on prototyping the model in Python and applying it to ADM data or on firmware programming of a model from the literature.

Aim

Investigation and implementation of a GRU in wearable contexts; Python prototyping and microcontroller programming

Data

Literature

Literature will be provided in the first meeting and the candidate is encouraged to further research relevant papers for this work.

Examination

Thesis report and final presentation.

Contact

David Kopyto

• Job title: Researcher
• Address:
  
  Henkestraße 91, Haus 7, 1. OG
  91052 Erlangen
  Germany
• Phone number: +49 9131 85-23608
• Email: david.kopyto@fau.de

Category: Student Projects, Work or Study with us, Work or Study with us Archive, WS2020/21