arXiv subjects

Christina Imdahl

Publications and source records attributed to Christina Imdahl.

4 recordsLinked to original sources

arXiv2026

Zero-shot Generalization in Inventory Management: Train, then Estimate and Decide

Deploying deep reinforcement learning (DRL) in real-world inventory management presents challenges, including dynamic environments and uncertain problem parameters, e.g. demand and lead time distributions. These challenges highlight a research gap, suggesting a need for a unifying framework to model and solve sequential decision-making under parameter uncertainty. We address this by exploring an underexplored area of DRL for inventory management: training generally capable agents (GCAs) under zero-shot generalization (ZSG). Here, GCAs are advanced DRL policies designed to handle a broad range of sampled problem instances with diverse inventory challenges. ZSG refers to the ability to successfully apply learned policies to unseen instances with unknown parameters without retraining. We propose a unifying Super-Markov Decision Process formulation and the Train, then Estimate and Decide (TED) framework to train and deploy a GCA tailored to inventory management applications. The TED framework consists of three phases: training a GCA on varied problem instances, continuously estimating problem parameters during deployment, and making decisions based on these estimates. Applied to periodic review inventory problems with lost sales, cyclic demand patterns, and stochastic lead times, our trained agent, the Generally Capable Lost Sales Network (GC-LSN) consistently outperforms well-known traditional policies when problem parameters are known. Moreover, under conditions where demand and/or lead time distributions are initially unknown and must be estimated, we benchmark against online learning methods that provide worst-case performance guarantees. Our GC-LSN policy, paired with the Kaplan-Meier estimator, is demonstrated to complement these methods by providing superior empirical performance.

cs.LG↗

arXiv2025

Strategic Selection of Remanufacturing Business Models: A Consumer Perception Perspective

As a key circular economy strategy, remanufacturing allows original equipment manufacturers (OEMs) to reduce waste by restoring used products to ``as-new'' conditions. This paper investigates an OEM's optimal remanufacturing business model by incorporating consumer perceptions into price and production quantity decisions. We analyze three alternative models: no remanufacturing, OEM in-house remanufacturing, and third-party remanufacturer (TPR) authorized remanufacturing. We extend the authorization with a two-part tariff contract and consider a stochastic market size. Through a numerical approach, we optimize price and quantity decisions based on consumer perceptions and develop a hierarchical decision roadmap to guide model selection. Our findings show that when consumer's perceived value of remanufactured products is high, OEM in-house remanufacturing is most profitable and reduces environmental impacts, but generally leads to a market dominated by remanufactured products. In contrast, when consumer's perceived value of remanufactured products is moderate and TPR remanufacturing significantly increases the perceived value of new products, the TPR-authorized remanufacturing is most profitable. It typically boosts total market sales, but accordingly increases environmental impacts. In addition, sensitivity analysis indicates that two-part authorization contracts are more advanced in meeting stringent environmental requirements than one-part contracts. Incorporating market size stochasticity enhances system profitability while keeping environmental impacts within a limited scope.

math.OC↗

arXiv2025

Deep Controlled Learning for Inventory Control

The application of Deep Reinforcement Learning (DRL) to inventory management is an emerging field. However, traditional DRL algorithms, originally developed for diverse domains such as game-playing and robotics, may not be well-suited for the specific challenges posed by inventory management. Consequently, these algorithms often fail to outperform established heuristics; for instance, no existing DRL approach consistently surpasses the capped base-stock policy in lost sales inventory control. This highlights a critical gap in the practical application of DRL to inventory management: the highly stochastic nature of inventory problems requires tailored solutions. In response, we propose Deep Controlled Learning (DCL), a new DRL algorithm designed for highly stochastic problems. DCL is based on approximate policy iteration and incorporates an efficient simulation mechanism, combining Sequential Halving with Common Random Numbers. Our numerical studies demonstrate that DCL consistently outperforms state-of-the-art heuristics and DRL algorithms across various inventory settings, including lost sales, perishable inventory systems, and inventory systems with random lead times. DCL achieves lower average costs in all test cases while maintaining an optimality gap of no more than 0.2\%. Remarkably, this performance is achieved using the same hyperparameter set across all experiments, underscoring the robustness and generalizability of our approach. These findings contribute to the ongoing exploration of tailored DRL algorithms for inventory management, providing a foundation for further research and practical application in this area.

cs.LG↗

arXiv2017

Möbius Moduli for Fingerprint Orientation Fields

We propose a novel fingerprint descriptor, namely Möbius moduli, measuring local deviation of orientation fields (OF) of fingerprints from conformal fields, and we propose a method to robustly measure them, based on tetraquadrilaterals to approximate a conformal modulus locally with one due to a Möbius transformation. Conformal fields arise by the approximation of fingerprint OFs given by zero pole models, which are determined by the singular points and a rotation. This approximation is very coarse, e.g. for fingerprints with no singular points (arch type), the zero-pole model's OF has parallel lines. Quadratic differential (QD) models, which are obtained from zero-pole models by adding suitable singularities outside the observation window, approximate real fingerprints much better. For example, for arch type fingerprints, parallel lines along the distal joint change slowly into circular lines around the nail furrow. Still, QD models are not fully realistic because, for example along the central axis of arch type fingerprints, ridge line curvatures usually first increase and then decrease again. It is impossible to model this with QDs, which, due to complex analyticity, also produce conformal fields only. In fact, as one of many applications of the new descriptor, we show, using histograms of curvature and conformality index (log of the absolute value of the Möbius modulus), that local deviation from conformality in fingerprints occurs systematically at high curvature which is not reflected by state of the art fingerprint models as are used, for instance, in the well known synthetic fingerprint generation tool SFinGe and these differences robustely discriminate real prints from SFinGe's synthetic prints.

stat.AP↗