Background and purpose
Federated learning (FL) has achieved a lot of attention in the past few years. FL is a distributed machine learning framework that promises efficiency and privacy benefits in settings where data is distributed among many clients. Although federated learning shows significant promise as a key solution when data cannot be shared, current methods show limited privacy properties and have shortcomings when applied to common real-world scenarios.

Given weights for a neural network, we want to minimize the sum of the expected loss over all clients. A central server is coordinating training between the clients. The most prevalent algorithm to solve this optimization problem is federated averaging. Each client has its own client model, which is trained on a local dataset. When all clients have completed the training, their weights are sent to the central server where they are aggregated into a global model using layer-wise averaging.

Most existing work on FL has been done on supervised learning, which requires that clients have labels associated with each training data point. Extending FL to unsupervised and semi-supervised settings where we have no, or very few labels, to train on present interesting and open challenges. In real world scenarios, there often is a high cost to label data, requiring some expert knowledge. Furthermore, client data is considered private and outsourcing this to some external party for labelling would not be possible. Therefore, client data in real world applications is often partly or completely unlabeled. However, if some labeled data is available at the central server, this could be utilized. This motivates research on federated learning in unsupervised, semi-supervised and self-supervised settings.

Thesis Description
In this master thesis, you will work on semi-supervised federated learning, and investigate methods to learn useful deep learning models. You will work in close collaboration with our deep learning research group in Gothenburg. The work requires skilled students within machine learning and statistical inference. You will be expected to do a literature study in order to get familiar with what the field looks like today, and then start with simpler models and eventually extend or develop upon more advanced solutions.

Supervisor: Edvin Listo Zec

Start date: Early 2022

Credits: 30 ECTS (högskolepoäng).

Who are you?
We expect you to have required skills:

  • Experience of implementing machine learning models.
  • Courses in mathematical statistics, probability theory or similar.
  • Programming skills. Preferably with some experience of relevant frameworks such as Pytorch or Tensorflow.

Welcome with your application!
For questions and more information, please contact recruiting manager Edvin Listo Zec, 0737200960. Candidates are encouraged to send in their application as soon as possible. Suitable applicants will be interviewed as applications are received. Last day for application is 31th of October.

Tillträde Enligt överenskommelse
Ort Göteborg
Län Västra Götalands län
Land Sverige
Referensnummer 2021/468
Kontakt
  • Edvin Listo Zec, 0737200960
Sista ansökningsdag 2021-11-07

Tillbaka till lediga jobb