Summary of Gas: Generative Auto-bidding with Post-training Search, by Yewen Li et al.
GAS: Generative Auto-bidding with Post-training Search
by Yewen Li, Shuai Mao, Jingtong Gao, Nan Jiang, Yunjian Xu, Qingpeng Cai, Fei Pan, Peng Jiang, Bo An
First submitted to arxiv on: 22 Dec 2024
Categories
- Main: Artificial Intelligence (cs.AI)
- 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 Generative auto-bidding, a new trend in facilitating online advertising, generates bids based on adjustable conditions using models like transformers and diffusers. However, these generative models suffer from low-quality data, leading to poor performance, especially in long sequential decision-making. To address this, we propose the Generative Auto-bidding scheme (GAS) using post-training search, which refines a base policy model’s output and adapts to various preferences. GAS employs weak-to-strong search alignment with small critics trained for different preferences and an MCTS-inspired search to refine the model’s output. Additionally, we provide a fine-tuning method for high-frequency preference scenarios considering computational efficiency. Experimental results on real-world datasets and online A/B tests demonstrate GAS’s effectiveness, achieving significant improvements (e.g., 1.554% increment of target cost) compared to existing methods. |
| Low | GrooveSquid.com (original content) | Low Difficulty Summary Imagine trying to find the perfect ad for a product without having to search through lots of options yourself. That’s what auto-bidding does – it helps advertisers by automatically choosing which ads to show and when. But, there are some problems with this system, like not getting good enough data or not being able to adapt to different preferences. To solve these issues, researchers came up with a new way called Generative Auto-bidding (GAS). GAS uses special models that can learn from data and adjust to fit different preferences. The results show that GAS is much better than the old ways, making it a big improvement for advertisers. |
Keywords
» Artificial intelligence » Alignment » Fine tuning
