STW: Research Assistant

We are searching for a research assistant who will work on:

• Building graphs from non-graph datasets using k-nearest neighbors (kNN), mutual-kNN, weighted b-matching, and learned similarity/edge scores.
• Comparing graph-based methods (e.g., label propagation, GCN/GraphSAGE) against non-graph baselines such as logistic regression, MLP, and tree-based models.
• Running controlled experiments across text, image, and tabular datasets while matching graph density and keeping features/data splits fixed.
• Analyzing not only accuracy/F1, but also graph structure: degree distribution, hubs, connectivity, homophily, and neighborhood diversity.
• Developing a reproducible benchmark and identify when graph construction helps, when it hurts, and why.

Why this project is interesting

The project sits at the intersection of graph algorithms and modern machine learning. We will study a basic question: how should the graph be constructed when it is not given? The results can provide preliminary evidence for a broader research program on graph discovery and may lead to a publication if the findings are sufficiently strong.

Requirements:

At least 3rd year undergraduate or higher 
Strong Python programming and data structures/algorithms
Basic machine learning
Comfort running experiments
Curiosity about graphs/GNNs
Prior GNN experience is helpful, not required
Able to read and write an English text document and understand it
Basic computer knowledge of working with Spreadsheet or similar tools