This page organizes papers related to artificial intelligence published around the world. This page is summarized using Google Gemini and is operated on a non-profit basis. The copyright of the paper belongs to the author and the relevant institution. When sharing, simply cite the source.
Automated bidding is a crucial tool for improving advertiser advertising performance. AI-based bidding (AIGB), which learns a conditional generative planner from offline data, has demonstrated superior performance to existing offline reinforcement learning (RL)-based automated bidding methods. However, existing AIGB methods face performance bottlenecks due to limitations in static offline datasets. To address these issues, this paper proposes AIGB-Pearl (Planning with Evaluator via RL), a novel method that integrates generative planning and policy optimization. The core of AIGB-Pearl is to construct a trajectory estimator that evaluates generation quality and to design a provably KL-Lipschitz constraint score maximization method to ensure safe and efficient generalization beyond offline datasets. Furthermore, we develop a practical algorithm that integrates synchronous coupling techniques to ensure model regularity for the proposed method. Extensive experiments on simulations and real-world advertising systems demonstrate the state-of-the-art performance of the proposed method.
Takeaways, Limitations
•
Takeaways:
◦
AIGB-Pearl integrates generative planning and policy optimization to overcome the limitations of static datasets and improve performance of existing AIGB methods.
◦
The KL-Lipschitz constraint score maximization method ensures safe and efficient generalization beyond offline datasets.
◦
The practicality of the proposed method was enhanced by securing model regularity through synchronous coupling technology.
◦
State-of-the-art performance was demonstrated through simulations and real-world advertising system experiments.
•
Limitations:
◦
There is no specific mention of Limitations in the paper (this may be revealed through further research).
