Summary of Improving Sentence Embeddings with Automatic Generation Of Training Data Using Few-shot Examples, by Soma Sato et al.
Improving Sentence Embeddings with Automatic Generation of Training Data Using Few-shot Examples
by Soma Sato, Hayato Tsukagoshi, Ryohei Sasano, Koichi Takeda
First submitted to arxiv on: 23 Feb 2024
Categories
- Main: Computation and Language (cs.CL)
- Secondary: Machine Learning (cs.LG)
GrooveSquid.com Paper Summaries
GrooveSquid.com’s goal is to make artificial intelligence research accessible by summarizing AI papers in simpler terms. Each summary below covers the same AI paper, written at different levels of difficulty. The medium difficulty and low difficulty versions are original summaries written by GrooveSquid.com, while the high difficulty version is the paper’s original abstract. Feel free to learn from the version that suits you best!
| Summary difficulty | Written by | Summary |
|---|---|---|
| High | Paper authors | High Difficulty Summary Read the original abstract here |
| Medium | GrooveSquid.com (original content) | Medium Difficulty Summary The paper explores ways to improve sentence embeddings without relying on large, manually annotated natural language inference (NLI) datasets. Currently, a decoder-based model called PromptEOL achieves high performance on semantic textual similarity tasks, but it requires fine-tuning using such datasets. The researchers aim to automate this process by generating an NLI dataset with a large language model and then fine-tuning PromptEOL using this automatically generated data. They investigate methods for generating suitable few-shot learning examples and evaluate their approach on STS tasks, demonstrating that it outperforms existing models in settings without large annotated datasets. |
| Low | GrooveSquid.com (original content) | Low Difficulty Summary The paper is about finding new ways to make computers better at understanding sentences without needing lots of labeled training data. Right now, a special computer model called PromptEOL does this job really well, but only if it’s trained using extra data that people have to label. The researchers want to change this by letting the computer itself generate some of this training data and then use it to improve its sentence- understanding abilities. |
Keywords
* Artificial intelligence * Decoder * Few shot * Fine tuning * Inference * Large language model
