In this paper, we investigate the adoption of federated learning approach in the domain of NLP requiring sensitive data. For this purpose, we have developed a federated learning infrastructure that performs training on remote devices without the need to share data. We demonstrate the usability of ...
This benchmark coming from theAllen Institute for AIseeks to standardize the patterns seen in the NLP few-shot literature, such as The principles upon which this benchmark is built are: diversity of transfer types, variable number of shots and classes, unbalanced training sets, textual labels, ...