Summary of Uncertainty-penalized Direct Preference Optimization, by Sam Houliston et al.
Uncertainty-Penalized Direct Preference Optimization
by Sam Houliston, Alizée Pace, Alexander Immer, Gunnar Rätsch
First submitted to arxiv on: 26 Oct 2024
Categories
- Main: Machine Learning (cs.LG)
- Secondary: Artificial Intelligence (cs.AI); Machine Learning (stat.ML)
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 new approach to aligning large language models (LLMs) with human preferences in content, style, and presentation. Current methods like reinforcement learning from human feedback (RLHF) and direct preference optimization (DPO) are prone to overoptimizing proxy rewards. To address this issue, the authors develop a pessimistic framework for DPO that incorporates preference uncertainty penalization schemes. These schemes correct the loss function by attenuating the gradient for uncertain samples. The proposed method is evaluated on the GPT2 Medium model using the Anthropic-HH dataset and demonstrates improved performance compared to vanilla DPO. |
| Low | GrooveSquid.com (original content) | Low Difficulty Summary The paper aims to make large language models (LLMs) more human-like by matching their content, style, and presentation with what people like. Right now, methods like giving feedback or optimizing preferences can be tricked into doing the wrong thing. The authors fix this problem by adding a new step that makes sure the model doesn’t get too good at doing the wrong thing. |
Keywords
» Artificial intelligence » Loss function » Optimization » Reinforcement learning from human feedback » Rlhf
