Abstract
In this paper we address the problem of fine-tuned text generation with a limited computational budget. For that, we use a well-performing text generative adversarial network (GAN) architecture - Diversity-Promoting GAN (DPGAN), and attempted a drop-in replacement of the LSTM layer with a self-attention-based Transformer layer in order to leverage their efficiency. The resulting Self-Attention DPGAN (SADPGAN) was evaluated for performance, quality and diversity of generated text and stability. Computational experiments suggested that a transformer architecture is unable to drop-in replace the LSTM layer, under-performing during the pre-training phase and undergoing a complete mode collapse during the GAN tuning phase. Our results suggest that the transformer architecture need to be adapted before it can be used as a replacement for RNNs in text-generating GANs.- Anthology ID:
- 2021.ranlp-1.21
- Volume:
- Proceedings of the International Conference on Recent Advances in Natural Language Processing (RANLP 2021)
- Month:
- September
- Year:
- 2021
- Address:
- Held Online
- Editors:
- Ruslan Mitkov, Galia Angelova
- Venue:
- RANLP
- SIG:
- Publisher:
- INCOMA Ltd.
- Note:
- Pages:
- 173–181
- Language:
- URL:
- https://aclanthology.org/2021.ranlp-1.21
- DOI:
- Cite (ACL):
- Kevin Blin and Andrei Kucharavy. 2021. Can the Transformer Be Used as a Drop-in Replacement for RNNs in Text-Generating GANs?. In Proceedings of the International Conference on Recent Advances in Natural Language Processing (RANLP 2021), pages 173–181, Held Online. INCOMA Ltd..
- Cite (Informal):
- Can the Transformer Be Used as a Drop-in Replacement for RNNs in Text-Generating GANs? (Blin & Kucharavy, RANLP 2021)
- PDF:
- https://preview.aclanthology.org/corrections-2024-07/2021.ranlp-1.21.pdf