02 oct
|
Universitat Politècnica De Catalunya (Upc)- Barcelonatech
|
España
02 oct
Universitat Politècnica De Catalunya (Upc)- Barcelonatech
España
Organisation/Company Universitat Politècnica de Catalunya (UPC)- BarcelonaTECH Research Field Engineering » Computer engineering Engineering » Communication engineering Mathematics » Applied mathematics Physics » Computational physics Researcher Profile First Stage Researcher (R1) Positions PhD Positions Application Deadline 5 Oct 2026 - 23:00 (Europe/Madrid) Country Spain Type of Contract Temporary Job Status Full-time Hours Per Week 37.5 Offer Starting Date 10 Oct 2026 Is the job funded through the EU Research Framework Programme? Not funded by a EU programme Reference Number 10267010007 Is the Job related to staff position within a Research Infrastructure? No
Offer Description
Mission:
Develop a doctoral thesis on the mechanistic interpretability of graph neural networks (GNN).
Functions to be developed:
- Build and openly publish a collection of GNNs trained on tasks with known answers (minimum paths, flows, spanning trees), to validate interpretability methods.
- Adapt to GNNs the interpretability tools created for large language models, such as sparse autoencoders,
to locate the variables that the network represents internally in nodes and edges.
- Design causal intervention experiments (e.g. activation patching) to check which of these variables the network actually uses, and choose the most compact description that remains faithful to its behavior.
- Apply the methods to real GNNs that act as digital twins of communications networks (such as RouteNet) and other sectors (energy, traffic, agri-food chain).
- Publish the results in top-level conferences and journals on machine learning and networks, release the code in open access and actively participate in the international interpretability community.
- Advanced programming in Python, PyTorch, PyTorch Geometric or DGL.
- Version control (git), Linux and model training on GPU (reproducible experiments).
Other requirements to consider:
- Curiosity to understand how AI models work internally, ability to work in
📌 Researcher in training to perform AI interpretability. (España)
🏢 Universitat Politècnica De Catalunya (Upc)- Barcelonatech
📍 España