Summary of Harmonic: Harnessing Llms For Tabular Data Synthesis and Privacy Protection, by Yuxin Wang et al.
HARMONIC: Harnessing LLMs for Tabular Data Synthesis and Privacy Protection
by Yuxin Wang, Duanyu Feng, Yongfu Dai, Zhengyu Chen, Jimin Huang, Sophia Ananiadou, Qianqian Xie, Hao Wang
First submitted to arxiv on: 6 Aug 2024
Categories
- Main: Machine Learning (cs.LG)
- Secondary: Artificial Intelligence (cs.AI); Computation and Language (cs.CL); Cryptography and Security (cs.CR)
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 paper tackles the challenge of obtaining tabular data from sensitive domains, a crucial step in advancing deep learning. Despite the emergence of Large Language Models (LLMs), generating realistic and privacy-preserving synthetic tabular data remains an urgent issue. The authors introduce HARMONIC, a framework for tabular data generation and evaluation that leverages LLMs with fine-tuning to produce high-quality synthetic data while preserving privacy. The approach uses the k-nearest neighbors algorithm to construct an instruction fine-tuning dataset, which trains LLMs to remember data relationships rather than the data itself, reducing privacy risks. The paper also proposes specific privacy risk metrics (DLT) and performance evaluation metrics (LLE) for evaluating synthetic data generation and downstream LLM tasks. Experiments show that HARMONIC achieves equivalent performance to existing methods with better privacy. |
| Low | GrooveSquid.com (original content) | Low Difficulty Summary This paper is about creating fake tabular data that’s just as good as real data, but without revealing private information. This is important because sometimes we need data from places where it’s not okay to get the real thing. The authors used a special kind of AI called Large Language Models (LLMs) to generate this synthetic data. They made sure the LLMs learned how to create realistic connections between different pieces of data, rather than memorizing the actual data itself. This helps keep private information safe. The paper also came up with new ways to measure how well the synthetic data works and how much privacy it preserves. |
Keywords
» Artificial intelligence » Deep learning » Fine tuning » Synthetic data
