Fig. 4
From: Probabilistic metabolite annotation using retention time prediction and meta-learned projections
Overview of the meta-learning approach to RTs projection. 1) The meta-tasks consist of creating projection functions mapping predicted RTs to experimental RTs in several Chromatographic Method (CM)s. Each target Chromatographic Method (CM) is a different meta-task. 2) During meta-learning a prior distribution \(p_{\theta }(f)\) on the projection functions is learned. This prior contains the learned projection functions in the meta-tasks (shown in color), and also any other function with similar properties to those observed in the dataset (shown in gray). 3) To solve a target-task, the prior distribution \(p_{\theta }(f)\) is updated with new evidence provided by the target training set, resulting in the so-called posterior distribution. 4) The posterior distribution is used to evaluate the performance on a target test set