Devanapalli, RS and Devi, VS (2018) Transfer learning using progressive neural networks and NMT for classification tasks in NLP. In: 25th International Conference on Neural Information Processing, ICONIP 2018, 13 - 16 December 2018, Siem Reap, pp. 188-197.
Full text not available from this repository.Abstract
Recently neural networks are obtaining state of the art results on many NLP tasks like sentiment classification, machine translation, etc. However one of the drawbacks of these techniques is that they need large amounts of training data. Even though there is a lot of data being generated everyday, not all tasks have large amounts of data. One possible solution when data is not sufficient is using transfer learning techniques. In this paper, we explored methods of transfer learning (or sharing the parameters) between different tasks so that the performance on the low data resource tasks is improved. We have first tried to replicate the prior results of transfer learning in semantically related tasks. When we have semantically different tasks, we tried using Progressive Neural Networks. We also experimented on sharing the encoder from neural machine translator to classification tasks.
| Item Type: | Conference Paper |
|---|---|
| Publication: | Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics) |
| Publisher: | Springer Verlag |
| Additional Information: | The copyright for this article belongs to the Springer Verlag. |
| Keywords: | Natural language processing systems; Signal encoding, Classification tasks; Data resources; Large amounts of data; Machine translations; Neural machine translator encoder; Sentiment classification; State of the art; Transfer learning, Classification (of information) |
| Department/Centre: | Division of Mechanical Sciences > Mechanical Engineering |
| Date Deposited: | 19 Aug 2022 04:44 |
| Last Modified: | 19 Aug 2022 04:44 |
| URI: | https://eprints.iisc.ac.in/id/eprint/75972 |
Actions (login required)
 |
View Item |
