TL;DR我们提出了 PLATO 方法,该方法通过使用描述输入特征的辅助知识图来规范多层感知器(MLP),在 d >> n 的表格数据上实现了强大的性能
Abstract
machine learning models exhibit strong performance on datasets with abundant
labeled samples. However, for tabular datasets with extremely high
$d$-dimensional features but limited $n$ samples (i.e. $d \gg n$), m
Knowledge-Enhanced Pre-trained Language Models improve downstream NLP tasks in closed domains by injecting knowledge facts from Knowledge Graphs using the proposed KANGAROO framework that captures implicit graph structure and employs data augmentation for better performance.