Papers by Maruan Al-Shedivat
Knowledge-Aware Meta-learning for Low-Resource Text Classification (2021.emnlp-main)
Copied to clipboard
| Challenge: | Existing meta-learning techniques may not be well-suited to testing tasks when they are not well-supported by training tasks. |
| Approach: | They propose to use a meta-learning algorithm to add representations for each sentence learned from the extracted sentence-specific knowledge graph. |
| Outcome: | The proposed model is able to represent each sentence learned from the extracted knowledge graph under supervised adaptation and unsupervised adaptation settings. |
Consistency by Agreement in Zero-Shot Neural Machine Translation (N19-1)
Copied to clipboard
| Challenge: | In this paper, we focus on zero-shot generalization—a challenging setup that tests models on translation directions they have not been optimized for at training time. |
| Approach: | They propose a method that allows for a consistent agreement-based training method that encourages the model to produce equivalent translations of parallel sentences in auxiliary languages. |
| Outcome: | The proposed model improves on public zero-shot translation benchmarks without loss of performance on supervised translation directions. |
Progressive Generation of Long Text with Pretrained Language Models (2021.naacl-main)
Copied to clipboard
| Challenge: | Existing methods for "long" text generation are limited to outputs of 50-200 tokens . however, our proposed ProGen generates coherent long passages of text in a progressive manner . |
| Approach: | They propose a method for generating coherent long passages of text in a progressive manner . they first produce domain-specific content keywords and then refine them into complete passages . human evaluation validates that their proposed generation is more coherent . |
| Outcome: | The proposed method produces domain-specific content keywords and refines them into complete passages in multiple stages. |