Summary of Personalized Federated Instruction Tuning Via Neural Architecture Search, by Pengyu Zhang et al.
Personalized Federated Instruction Tuning via Neural Architecture Search
by Pengyu Zhang, Yingbo Zhou, Ming Hu, Junxian Feng, Jiawen Weng, Mingsong Chen
First submitted to arxiv on: 26 Feb 2024
Categories
- Main: Machine Learning (cs.LG)
- Secondary: None
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 This research paper proposes a novel framework called Personalized Federated Instruction Tuning (PerFIT) to overcome the limitations of Federated Instruction Tuning (FIT). PerFIT allows each client to search for a personalized architecture by expanding the trainable parameter space, followed by pruning the parameters to maintain the same number as the original model. This approach enables personalized instruction fine-tuning within expanded parameter spaces while preserving the same number of trainable parameters. The framework also utilizes personalized parameter-wise aggregation to release the abilities of heterogeneous computational resources and enhance performance on local data. Experimental results with multiple language models (LLMs) in non-IID scenarios demonstrate that PerFIT can achieve a 23% decrease in perplexity compared to state-of-the-art FIT methods. |
| Low | GrooveSquid.com (original content) | Low Difficulty Summary PerFIT is a new way to help many people work together on a project without sharing their private information. The problem was that each person had different data and computers, which made it hard to adapt the instructions to their specific needs. PerFIT lets each person find their own best approach by looking at a wider range of options and then simplifying them back down to what they started with. This helps people with stronger computers or more unique data get better results. The researchers tested this method on several language models and found that it worked really well, reducing the error rate by 23% compared to other methods. |
Keywords
* Artificial intelligence * Fine tuning * Instruction tuning * Perplexity * Pruning
