Reproduce “Using Sequences of Life-events to Predict Human Lives” on OpenResearch
Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.
Universal app for Apple silicon and Intel. Requires macOS 11 or later.
About this paper
Germans Savcisens, Tina Eliassi-Rad, Lars Kai Hansen, Laust Mortensen, Lau Lilleholt, Anna Rogers, Ingo Zettler, Sune Lehmann
`life2vec` is a transformer-based framework that applies natural language processing techniques to comprehensive, nation-scale sequences of human life-events from Danish national registers. It successfully predicts outcomes like early mortality and personality nuances, outperforming state-of-the-art baselines by up to 11% while also providing interpretable insights into learned event relationships.
Key insights
- Human life trajectories can be effectively modeled as "sequences of events" using transformer architectures, similar to natural language.
- The model learns highly structured and robust concept embeddings that reveal meaningful global relationships (e.g., by event type) and fine-grained semantic proximities (e.g., clustering of related diagnoses or job types).
- Task-conditioned person embeddings are generated, which organize individuals along a continuum related to the predicted outcome, with interpretable directions revealed by tools like TCAV.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.