Decision-Focused Forecasting: Decision Losses for Multistage Optimisation

2024-05-23

arXiv.org

AI Summary

Many real-world decision-making problems exhibit multi-stage structures, temporal dependencies, and intertemporal effects—characteristics inadequately addressed by prevailing decision-focused learning methods, which predominantly assume single-stage settings. To bridge this gap, we propose the first decision-oriented predictive framework explicitly designed for multi-stage optimization: it jointly trains predictive models and dynamic decision policies in an end-to-end differentiable manner, explicitly encoding temporal decision dependencies via differentiable optimization. Furthermore, we introduce an implicit multi-layer recurrent architecture that captures feedback from state trajectories to predictions, and theoretically establish its gradient-based self-correcting mechanism for prediction bias. Evaluated on an energy storage arbitrage task, our method substantially outperforms both decoupled predict-then-optimize baselines and single-stage counterparts, achieving a 12.7% improvement in long-horizon cumulative reward.

Technology Category

• Search and Optimization: Learning to Search
• Reasoning under Uncertainty: Sequential Decision Making
• Intelligent Robots: Learning & Optimization for ROB

Application Category

• Search and Retrieval-Augmented AI: Web learning to rank, online learning, and counterfactual learning for ranking
• User Modeling, Personalization and Recommendation: Fairness-aware retrieval and ranking
• Graph Algorithms and Modeling for the Web: Graph neural networks and deep learning approaches for Web-related graphs

Abstract

Decision-focused learning has emerged as a promising approach for decision making under uncertainty by training the upstream predictive aspect of the pipeline with respect to the quality of the downstream decisions. Most existing work has focused on single stage problems. Many real-world decision problems are more appropriately modelled using multistage optimisation as contextual information such as prices or demand is revealed over time and decisions now have a bearing on future decisions. We propose decision-focused forecasting, a multiple-implicitlayer model which in its training accounts for the intertemporal decision effects of forecasts using differentiable optimisation. The recursive model reflects a fully differentiable multistage optimisation approach. We present an analysis of the gradients produced by this model showing the adjustments made to account for the state-path caused by forecasting. We demonstrate an application of the model to an energy storage arbitrage task and report that our model outperforms existing approaches.

Problem

• Extends decision-focused learning to multistage optimization problems
• Addresses intertemporal effects of sequential decisions in forecasting
• Improves performance in energy storage and portfolio optimization

Innovation

• Differentiable multistage optimisation architecture
• Recurrent differentiable optimisation approach
• Gradient adjustments for intertemporal decision effects