General Track Winner (MingGeZheJiu) Presentation

In this auto-bidding competition, we develop a time-varying, risk-avoiding linear bidding strategy that achieves a score of 0.5635 in the final rounds. Building on prior work, we first establish a general bidding model for Target CPA bidding. By leveraging the cyclic and temporal patterns of pValue, we introduce time-varying parameterization. Finally, we optimize the bidding parameters in a simulated offline environment and incorporate online feedback for fine-tuning, ultimately reaching a score of 0.5635.