Asymptotic Bayes risk of semi-supervised multitask learning on Gaussian mixture
This work addresses performance analysis in multitask learning for statistical models, but it is incremental as it applies existing statistical physics methods to a specific setting.
The paper tackled the problem of semi-supervised multitask learning on a Gaussian mixture model by computing the asymptotic Bayes risk to analyze task similarity and performance gains from joint learning, deriving a Bayes optimal algorithm for the supervised case.
The article considers semi-supervised multitask learning on a Gaussian mixture model (GMM). Using methods from statistical physics, we compute the asymptotic Bayes risk of each task in the regime of large datasets in high dimension, from which we analyze the role of task similarity in learning and evaluate the performance gain when tasks are learned together rather than separately. In the supervised case, we derive a simple algorithm that attains the Bayes optimal performance.