Summary of Self-consistency Preference Optimization, by Archiki Prasad et al.
Self-Consistency Preference Optimization
by Archiki Prasad, Weizhe Yuan, Richard Yuanzhe Pang, Jing Xu, Maryam Fazel-Zarandi, Mohit Bansal, Sainbayar Sukhbaatar, Jason Weston, Jane Yu
First submitted to arxiv on: 6 Nov 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 paper introduces Self-Consistency Preference Optimization (ScPO), a novel method that extends the concept of self-consistency from inference time to training models. ScPO iteratively trains consistent answers to be preferred over inconsistent ones on unsupervised new problems, leading to large improvements in complex reasoning tasks such as GSM8K and MATH. The approach bridges the gap with supervised training using gold answers or preferences, and combining it with standard supervised learning further improves results. ScPO also finetunes language models like Llama-3 8B to outperform other models on tasks like ZebraLogic. |
| Low | GrooveSquid.com (original content) | Low Difficulty Summary The paper helps machines learn better by themselves without needing human help. It introduces a new way to train models, called Self-Consistency Preference Optimization (ScPO), which makes them prefer correct answers over incorrect ones. This improves how well the models do complex reasoning tasks, and even gets close to what humans can do with labeled data. The approach also helps fine-tune language models to be better at specific tasks. |
Keywords
» Artificial intelligence » Inference » Llama » Optimization » Supervised » Unsupervised
