Reproduce “Intermittent Demand Forecasting with Deep Renewal Processes” 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
Ali Caner Turkmen, Yuyang Wang, Tim Januschowski
This research from Boğaziçi University and AWS AI Labs introduces Deep Renewal Processes (DRPs), a framework combining renewal process theory with deep learning to forecast intermittent demand. DRPs utilize recurrent neural networks to dynamically model interdemand times and demand sizes, consistently outperforming traditional methods and providing improved uncertainty quantification on real-world datasets.
Key insights
- Intermittent demand forecasting can be formally viewed as a renewal process, where events are demand occurrences and interarrival times are periods between demands.
- Deep learning models, specifically recurrent neural networks (RNNs), can effectively capture dynamic conditional distributions of interdemand times and demand sizes, overcoming limitations of static distributional assumptions.
- A global modeling approach, training on multiple time series simultaneously, allows Deep Renewal Processes (DRPs) to learn common patterns and generalize well across diverse intermittent demand series.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.