TIP-Search: Time-Predictable Inference Scheduling for Market Prediction under Uncertain Load

TIP-Search introduces a time-predictable inference scheduling method for market prediction under uncertain load, focusing on delivering timely and accurate predictions. It utilizes a systems-replay controller, OCO-ACPO, and its extension SA-OCO-ACPO, achieving raw accuracy of 0.994 and timely accuracy of 0.991, while significantly improving deadline satisfaction metrics. This work is critical for practitioners as it enhances the reliability of real-time market predictions, balancing accuracy and deadline adherence in dynamic environments.