First-order meta-learning algorithms have been widely used in practice to learn initial model parameters that can be quickly adapted to new tasks due to their efficiency and effectiveness. However, existing studies find that meta-learner can overfit to some specific adaptation when we have heterogeneous tasks, leading to significantly degraded performance. In Natural Language Processing (NLP) applications, datasets are often diverse and each task has its unique characteristics. Therefore, to address the overfitting issue when applying first-order meta-learning to NLP applications, we propose to reduce the variance of the gradient estimator used in task adaptation. To this end, we develop a variance-reduced first-order meta-learning algorithm. The core of our algorithm is to introduce a novel variance reduction term to the gradient estimation when performing the task adaptation. Experiments on two NLP applications: few-shot text classification and multi-domain dialog state tracking demonstrate the superior performance of our proposed method.

This paper describes a corpus of nearly 10K French-Chinese aligned segments, produced by post-editing machine translated computer science courseware. This corpus was built from 2013 to 2016 within the PROJECT_NAME project, by native Chinese students. The quality, as judged by native speakers, is ad-equate for understanding (far better than by reading only the original French) and for getting better marks. This corpus is annotated at segment-level by a self-assessed quality score. It has been directly used as supplemental training data to build a statistical machine translation system dedicated to that sublanguage, and can be used to extract the specific bilingual terminology. To our knowledge, it is the first corpus of this kind to be released.