On this page you can find our offering for semester projects. Students who are interested in doing a project are encouraged to have a look at the Thesis & Project Guidelines from the MLO lab, where you will gain an understanding about what can be expected of us and what we expect from students.
Last Update: 26th January 2023
Please, apply via Google form. You will need to specify which project(s) you are interested in, why you are interested, and if you have any relevant experience in this area. To access the form you need to login with your EPFL email address.
We process students in two rounds. We collect applications for projects for the first round until 25th November 2022. We will get back to you about your application in the first week after the deadline. If we do not get back to you during the indicated period, this means that we unfortunately did not have space.
We will leave the form open for late applications. If by 8th January 2023 there are still projects open we will consider all applications submitted by then. We strongly recommend that you apply as soon as possible for best consideration. We expect many projects to be taken after the first round.
External students: Students who are not from EPFL should get in touch with the supervisors of the project(s) via email.
Early deadline: 25th November 2022
First Contact with Supervisors: 5th December 2022 - 9th December 2022
Late deadline: 8th January 2023
First Contact with Supervisors: 9th January 2023 - 13th January 2023
If you encounter any technical issue please get in touch with Laurent Girod.
Fully-Decentralized Learning (DL) has been proposed to circumvent the main limitations of Federated Learning (FL). In DL, there is no central server and users communicate with each other via peer-to-peer channels, drastically reducing bandwidth consumption. More importantly, this makes the underlying system more private as there is no central server with strong adversarial capabilities as in FL, right? Well, this is not the case.
In this project, you will deploy and run advanced privacy attacks (including gradient inversion) against real-world systems, proving that fully-decentralized learning protocols are still far from reaching the claimed security.
Collaborative learning allows distinct participants to train a joint model, without their local training data leaving the devices. See Google’s paper for an example. Such approach has gained a lot of attention recently because of its claimed efficiency, privacy, and security. However, those claims rely on a “good” setup phase, which includes the choice of graph connecting users, initial parameters, optimizer, and many more.
In this project, the student is expected to challenge the (sometimes not-so realistic) assumptions and design choices of existing setup designs, and explore their impact on the training process.
Further details This semester project is aimed at one MSc student (BSc can be considered as well). The student will work with Mathilde Raynal.
To understand how infectious diseases spread, researchers carry out experiments in which they study mobility and interaction patterns. These datasets have high value for science, but also encode a large amount of private information. How to quantify this information in order to decide whether the data can be published without harming the subjects of the experiment is an open problem.
In this project, our aim is to develop a privacy evaluation framework that helps epidemiology researchers to make informed decisions when publishing the data collected in their experiments. To this end, the student will have to develop algorithms to anonymize the data, and attacks to de-anonymize the datasets once we apply these algorithms. For these attacks the student will have to identify useful features to develop a machine learning model that can re-identify individuals and groups in epidemiological data.
Have you heard about the recent Lastpass’s breach? Apparently, users’ (encrypted) password vaults have been leaked. Researchers think that password vaults can be made “uncrackable” by relying on honey-encryption. Personally, I don’t buy it. Would you help me debunk that and similar claims?
If the answer is yes, this is the project for you. We are going to attack honey-encryption and honey-words systems, showing that those can offer only a false sense of security. The attacks are going to be deep-learning-based. So, knowledge about ML is mandatory.
Further details This semester project is aimed at one bachelor or master student. The student will work with Dario Pasquini.
Note: Other labs with cool projects on Machine Learning (we can always consider co-supervision):
In this project you will be implementing a privacy-friendly aid distribution system prototype for the International Committee of the Red Cross (ICRC). One of the key challenges in aid distribution is that due to fraud at different places, aid does not always end up with the people that need aid the most. This is problematic, because the total amount of resources available to the ICRC is limited.
We designed a system that preserves the privacy of recipients of aid while at the same time making fraud more difficult. In this system, each aid recipient will receive a smart card. In this project you will implement a prototype of this system, which includes a smart card implementation as well as the terminal software that will be used when registering recipients and when distributing aid.
Further details This semester project is aimed at one bachelor or master student. The student will work with Kasra EdalatNejad.
You will contribute to the implementation of a prototype system for multiparty end-to-end encrypted computation. Such systems enable multiple users to compute joint functions over their local (and potentially private) data, yet without revealing their input data to the other parties in plaintext. This can be achieved thanks to advanced cryptographic construction such as homomorphic encryption. We have designed and implemented a first version of this system in Go, and we are now making it ready for making it open-source.
Your tasks may include several of the following (to be defined according to interests of the interested candidate):
Further details This semester project is aimed at one MSc student. The student will work with Christian Mouchet.
The SPRING lab has been working with different organizations of Journalists to develop technologies that enable them to perform their work in a safe manner. Among the actions that can entail risk, digital interactions with sources and informants is one of the most sensitive.
In this project, we will work with journalists to build a tool to help them understand the risks that stem from having interactions through digital means. Once risks are understood we will design alternative communication means that mitigate those risks.
In this project, you will contribute towards improving the privacy and functionality of NYM’s anonymous communication mixnet. To ensure maximum impact of the project you will be working in close collaboration with NYM’s research and engineering teams.
Your tasks may include several of the following (to be defined according to interests of the interested candidate and NYM’s needs at the time of starting the project):
Further details This semester project is aimed at one MSc or PhD student. The student will work with Carmela Troncoso.