Student projects in the EPFL NeuroAI Lab should:
- contribute to the mission of the lab, i.e. to create accurate models of behavior and underlying neural mechanisms — these models let us make sense of the mind and brain, and enable new applications in the diagnosis and treatment of neural disorders
- contribute to ongoing projects by permanent members (PhDs and Postdocs), so that the results of the project will be useful long-term, and so that the student can make use of existing expertise.
- be aligned to the interests of the student with a realistic timeline so that meaningful progress with some closure can be made over the duration of the project.
The list of projects here is non-comprehensive. It is always a good idea to reach out to lab members whose research you find interesting and inquire if you can contribute to their projects. In either case, we ask that you apply via the form on the lab website. During interviews, we will share a more comprehensive list of projects with you that we are thinking about.
Learning trajectories in models and children
Jointly with Nina Kazanina and Denis Jabaudon.
(1 student)
Supervision
by Martin, Nina, and Denis. Due to limited PI time, the student is expected to be fairly independent already.
Project
Children acquire language on a remarkably consistent schedule: phoneme categories narrow before the first words, words before word combinations, and complex syntax years later. Today’s language models reach comparable competence, but do they get there the same way? This project treats model training as a developmental trajectory and asks whether the order and timing of landmark abilities coming online in artificial systems match those in children — evaluated with the same instruments pediatricians and cognitive scientists use, from vocabulary inventories and grammaticality judgments to looking-time and comprehension measures. Existing work has compared a handful of syntactic milestones in text-only models; we aim to go further along three axes that we think matter developmentally: models that learn their own units from raw input rather than inheriting a pre-trained tokenizer, joint learning across text, vision and audio rather than modalities grafted onto a mature language backbone, and constrained “genomic” initializations that ask how much structure must be built in before learning begins. The work combines training and densely checkpointing models from scratch with structural probing of what they represent, in collaboration with developmental neuroscientists and psycholinguists. Suitable for someone with a background in machine learning, computational linguistics, or cognitive science who wants to work at the interface — and who is interested in what the mismatches, as much as the matches, reveal about how language is learned.
Related work: https://aclanthology.org/2023.findings-acl.773/, https://aclanthology.org/2022.tacl-1.1/, https://proceedings.neurips.cc/paper_files/paper/2024/hash/8d987b2981388c99c7eab6095d1d29fd-Abstract-Datasets_and_Benchmarks_Track.html, https://www.science.org/doi/10.1126/science.adi1374, https://www.nature.com/articles/s41467-019-11786-6, https://arxiv.org/abs/2507.07955
Characterizing Neuronal Activity in Brain Organoids Across Neurodevelopmental Disorder Mutants
Project lead by Fides Zenk lab — please apply directly to [email protected]
(1 student)
Supervision: Fully supervised by an experienced postdoctoral researcher who will provide guidance
throughout the analysis.
Background
Brain organoids, miniature brain-like structures grown from human stem cells, offer an exciting model
to study how genetic mutations affect early brain development. Many mutations linked to
neurodevelopmental disorders (NDDs), such as autism or Rett syndrome, are thought to alter how
neurons communicate and form networks. In this project, you will analyze electrical recordings from
organoids carrying such mutations and use computational tools to identify what makes their activity
patterns distinctive.
Project Overview
The experimental data will already be in hand: multielectrode array (MEA) recordings from organoids
carrying NDD-associated mutations alongside healthy controls. Your role will be to dive into this rich
dataset and extract meaning from it.
The project has two main components:
1. Characterizing Neural Activity. You will process and explore the recordings, quantifying features of
network activity such as firing rates, burst patterns, and synchrony. The goal is to build an intuition
for how activity differs across genotypes and to produce clean, well-documented analyses.
2. Building Predictive & Interpretable Models. Using the extracted features, you will train machine
learning models to classify mutant versus wildtype organoids. Importantly, the focus won’t just be on
prediction accuracy, you’ll also ask why the model works: which features of neural activity are most
informative? This interpretability angle turns the classifier into a tool for biological discovery.
As an optional extension, depending on progress, you could explore how activity patterns change
across developmental timepoints, asking when mutant organoids first start to look different from
controls.
What You Will Learn
• Fundamentals of neural data analysis (spike sorting, time-series features)
• Practical machine learning for biological data (Python, scikit-learn)
• How to work at the interface of experiment and computation
• Scientific communication in an interdisciplinary setting
Support & Environment
You will be embedded in a collaborative environment between two labs and closely mentored by a
postdoc with direct experience generating and interpreting this data. This is a great opportunity to
develop both computational and neuroscientific thinking no prior organoid experience is required.