Advancing AI in Logistics: Our Latest Research on Time Series Transformers

The Chair of Logistics and Quantitative Methods at the University of Würzburg is proud to announce the acceptance of three of our research papers at prestigious international conferences in recent months. Our work focuses on leveraging Time Series Foundation Models and generative AI to revolutionize logistics and supply chain planning.

1. "In-context Quantile Regression for Multi-product Inventory Management using Time-series Transformers"
   Accepted as a poster at the NeurIPS 2024 Workshop on Time Series in the Age of Large Models in Vancouver.
   Read more
   For this paper, you can also watch our 5-minute video presenting all the major ideas and results.

2. "Predictive Inference Is Really Free with In-Context Learning"
   Accepted as a poster at the ICLR 2025 Workshop: Quantify Uncertainty and Hallucination in Foundation Models: The Next Frontier in Reliable AI in Singapore.
   Read more

3. "What Can We Learn from LLMs? Building a Foundation Model for Inventory Management"
   Accepted at the European Conference on Information Systems (ECIS) 2025 in Amman, Jordan.
   Read more

These papers showcase how transformer models—the same architecture used in advanced language models like ChatGPT and Gemini—can be trained on vast datasets of tens of thousands of products to generate significantly improved probabilistic forecasts compared to traditional machine learning methods. This advancement has the potential to greatly enhance planning processes, reduce costs, and minimize emissions in real-world applications. Furthermore, our findings highlight the model's capability to deliver accurate predictions for newly introduced products, offering additional practical advantages and further strengthening its impact on supply chain management.

Additionally, our research explores methods to quantify model uncertainty, a critical aspect for ensuring decision-makers can confidently rely on these models in practice.

Our work remains ongoing, and we are committed to further advancing our models’ capabilities, with plans to publish follow-up studies throughout the year.

Magnus Maichle, Kai Günder, Sohom Mukherjee, and Ivane Antonov attended the NeurIPS 2024 Workshop on Time Series in the Age of Large Models in Vancouver last December to present our work. This opportunity allowed us to engage with leading experts from Amazon, Google, Microsoft, and Salesforce, providing valuable insights that will shape our future research endeavors.

Stay tuned for more updates as we continue to push the boundaries of AI-driven logistics and supply chain management!