Summary of Ks-lottery: Finding Certified Lottery Tickets For Multilingual Language Models, by Fei Yuan et al.
KS-Lottery: Finding Certified Lottery Tickets for Multilingual Language Models
by Fei Yuan, Chang Ma, Shuai Yuan, Qiushi Sun, Lei Li
First submitted to arxiv on: 5 Feb 2024
Categories
- Main: Computation and Language (cs.CL)
- Secondary: Artificial Intelligence (cs.AI); 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 lottery ticket hypothesis suggests that within a randomly initialized neural network, there exist “winning tickets” that can be fine-tuned for better performance. This paper investigates whether these winning tickets exist in large language models (LLMs) during multilingual fine-tuning scenarios and proposes a method to identify them, called KS-Lottery. The approach utilizes the Kolmogorov-Smirnov Test to analyze parameter distribution shifts before and after fine-tuning. Notably, theoretical proofs demonstrate that KS-Lottery can find certified winning tickets in embedding layers, ensuring performance comparable to full fine-tuning LLMs. Experimental results on translation tasks show that KS-Lottery identifies a smaller set of parameters for fine-tuning, achieving similar performance as full fine-tuning LLMs. Surprisingly, fine-tuning just 18 tokens’ embeddings of LLaMA suffices to reach the desired fine-tuning translation performance. |
| Low | GrooveSquid.com (original content) | Low Difficulty Summary Large language models (LLMs) can be fine-tuned for better performance in multilingual scenarios. Researchers discovered that some parameters within the model are more important than others and proposed a method called KS-Lottery to find these “winning tickets”. The idea is to analyze how the model’s parameters change before and after fine-tuning. This helps identify which parts of the model are most useful for improving performance. The results show that by focusing on just a few key areas, we can achieve similar results as if we had fine-tuned the entire model. |
Keywords
* Artificial intelligence * Embedding * Fine tuning * Llama * Neural network * Translation
