Binning with nonparametric block models

Combining cluster assignments with learned embeddings

Joint work with Asst. Prof. Abdulkadir Çelikkanat on bridging clustering and representation learning for genomic data. The idea is to combine the hard cluster assignments produced by binners with learned continuous embeddings, drawing on nonparametric Bayesian block models — the Infinite Relational / Stochastic Block Model family — so that the number of groups need not be fixed in advance and structure can emerge from the data itself.