Summary of Inducing Generalization Across Languages and Tasks Using Featurized Low-rank Mixtures, by Chu-cheng Lin and Xinyi Wang and Jonathan H. Clark and Han Lu and Yun Zhu and Chenxi Whitehouse and Hongkun Yu
Inducing Generalization across Languages and Tasks using Featurized Low-Rank Mixtures
by Chu-Cheng Lin, Xinyi Wang, Jonathan H. Clark, Han Lu, Yun Zhu, Chenxi Whitehouse, Hongkun Yu
First submitted to arxiv on: 27 Feb 2024
Categories
- Main: Computation and Language (cs.CL)
- Secondary: Artificial Intelligence (cs.AI)
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 proposes a novel parameter-efficient fine-tuning (PEFT) method, Featurized Low-rank Mixtures (FLix), for adapting pretrained large language models (LLMs) to various downstream tasks across hundreds of human languages. The existing LoRA method suffers from suboptimal performance when dealing with diverse dataset mixtures due to aggressive parameter tying and negative interference among datasets. FLix addresses this issue by associating each unique dataset feature, such as the dataset’s language or task, with its own low-rank weight update parameters. This allows FLix to effectively accommodate diverse dataset mixtures and generalize better to unseen datasets. The proposed method demonstrates significant improvements over a variety of tasks for both supervised learning and zero-shot settings. |
| Low | GrooveSquid.com (original content) | Low Difficulty Summary The paper is about making machines understand many languages. It’s hard to teach a machine to do this because it needs lots of data and computing power. A new way to do this called Featurized Low-rank Mixtures (FLix) is proposed. FLix helps the machine learn from different types of data, like language or task, which makes it better at understanding many languages. This new method works well for both learning with labeled data and learning without any examples. |
Keywords
* Artificial intelligence * Fine tuning * Lora * Parameter efficient * Supervised * Zero shot
