[#05] Relaxing conditional independence assumptions in networks: the case of communities and reciprocity

She presented (see her slides) two variants of probabilistic generative models that relax the assumption of conditional independence: one specifies conditional probabilities and relies on a pseudo-likelihood approximation, while the other jointly model pairs of edges with exact 2-edge joint distributions.

These attempts in relaxing the assumption are reflected in improving the performance in edges prediction and recovering communities. Remarkably, these models are capable of generating synthetic networks that replicate the reciprocity values observed in real networks.

Martina showed us how she and her collaborators needed to build these models step by step, consciously choosing what can be done and what to leave for further research.

Sometimes, at the beginning of a new project, is usual to feel lost. Facing a new, general question to address always came with expectation and excitement about the many research possibilities. However, right after you start working, also a ton of small difficulties start to pop out - they are the price to pay for the research reward and the everyday meal of a scientist, regardless of the stage of your academic career.

The workload begins to increase and the general question branches out in an unpredictable way. Problems that are not solvable show up - at least in the short time of your PhD: so you need to move on. But this isn’t always easy … how to decide when to switch to something else and more accessible? Share your them on Bluesky!