Reproduce “Generative Auto-Bidding with Unified Modeling and Exploration” on OpenResearch
Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.
Universal app for Apple silicon and Intel. Requires macOS 11 or later.
About this paper
Mingming Zhang, Feiqing Zhuang, Na Li, Shengjie Sun, Xiaowei Chen, Junxiong Zhu, Fei Xiao, Keping Yang and 2 more
Guide, a generative auto-bidding framework, unifies the modeling of advertising dynamics with directed exploration and safety mechanisms. It achieves robust performance by combining a Decision Transformer for exploratory actions, an Inverse Dynamics Module for stable fallback actions, and a Q-value module for adaptive selection, leading to a 4.10% increase in Ad GMV and 3.52% in Ad ROI in online A/B tests on Taobao.
Key insights
- The unified modeling approach, simultaneously generating actions and predicting future states via a Decision Transformer and an Inverse Dynamics Module, enhances the understanding of environmental dynamics.
- A principled 'explore–safeguard–select' paradigm effectively reconciles the tension between discovering novel, high-performing bidding strategies and ensuring financial stability and risk mitigation.
- A two-stage training strategy (separate pre-training followed by joint training) for the DT and IDM provides improved stability and faster convergence compared to single-stage joint training.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.