Distribution Embedding Networks for Generalization from a Diverse Set of Classification Tasks
This addresses the challenge of scarce labeled data in related tasks for tabular-data classification, offering a more flexible meta-learning approach.
The paper tackles the problem of classification with small data by proposing Distribution Embedding Networks (DEN), which learn from diverse training tasks to generalize to unseen target tasks with heterogeneous input spaces, particularly for tabular data. The method outperforms existing approaches in synthetic and real tasks, as demonstrated in numerical studies.
We propose Distribution Embedding Networks (DEN) for classification with small data. In the same spirit of meta-learning, DEN learns from a diverse set of training tasks with the goal to generalize to unseen target tasks. Unlike existing approaches which require the inputs of training and target tasks to have the same dimension with possibly similar distributions, DEN allows training and target tasks to live in heterogeneous input spaces. This is especially useful for tabular-data tasks where labeled data from related tasks are scarce. DEN uses a three-block architecture: a covariate transformation block followed by a distribution embedding block and then a classification block. We provide theoretical insights to show that this architecture allows the embedding and classification blocks to be fixed after pre-training on a diverse set of tasks; only the covariate transformation block with relatively few parameters needs to be fine-tuned for each new task. To facilitate training, we also propose an approach to synthesize binary classification tasks, and demonstrate that DEN outperforms existing methods in a number of synthetic and real tasks in numerical studies.