Abstract
Despite the recognised potential of horizontal collaboration in logistics to reduce inefficiencies, and the increasing academic interest in this topic, in practice many initiatives fail. One of the main reasons for this failure is the poor strategy planning and governance organisation. This paper addresses this gap proposing a comprehensive conceptual framework to support the design and implementation of a common strategy for the stakeholders of such partnerships. The research employs qualitative methods, drawing on interviews and the case analysis of existent initiatives. The proposed framework involves the main phases of the strategic formulation, deciding the stakeholder engagement, strategic formulation, operational implementation, and business model elaboration. It serves as a road map for stakeholders to avoid common mistakes and accelerate the deployment of cooperative partnerships.

Abstract

Abstract
Recommender systems (RS) on platforms like YouTube are often evaluated as if they operate in a closed environment. In practice, however, user consumption patterns are also shaped by a broader ecosystem of external sources. This paper investigates how external link interactions influence RS behavior. We conducted a controlled experiment with three intervention timings and found that a single external link exerts an immediate and significant impact on YouTube's recommendations, an influence that decays but persists over time. These findings contribute to our understanding of how external interactions shape RS outputs and their subsequent impact on content diversity.

Abstract
Students in computing education increasingly use large language models (LLMs) such as ChatGPT. Yet, the role of LLMs in supporting cognitively demanding tasks, like deductive program verification, remains poorly understood. This paper investigates how students interact with an LLM when solving formal verification exercises in Dafny, a language that supports functional correctness by allowing programmers to write formal specifications and automatically verifying that the implementation satisfies the specification. We conducted a mixed-methods study with master's students enrolled in a formal methods course. Each participant completed two verification problems, one with access to a custom ChatGPT interface that logged all interactions and the other without. We identified strategies used by successful students and assessed the level of trust students place in LLMs. Our findings show that students perform significantly better when using ChatGPT; however, performance gains are tied to prompt quality. We conclude with practical recommendations for integrating LLMs into formal methods courses more effectively, including designing LLM-aware challenges that promote learning.

Abstract
The recent developments of modern probabilistic programming languages have enabled the combination of pattern recognition engines implemented by neural networks to guide inference over explanatory factors written as symbols in probabilistic programs. We argue that learning to invert fixed generative programs, instead of learned ones, places stronger restrictions on the representations learned by feature extraction networks, which reduces the space of latent hypotheses and enhances training efficiency. To empirically demonstrate this, we investigate a neurosymbolic object-centric representation learning approach that combines a slot-based neural module optimized via inference compilation to invert a prior generative program of scene generation. By amortizing the search over posterior hypotheses, we demonstrate that approximate inference using data-driven sequential Monte Carlo methods achieves competitive results when compared to state-of-the-art fully neural baselines while requiring several times fewer training steps. © 2026, Transactions on Machine Learning Research. All rights reserved.

Abstract
Stress is associated with many negative effects, including inadequate sleep, reduced learning and memory, and a higher risk of mental health conditions. Given these effects, it is important to explore effective strategies for stress management and intervention. One promising approach is the use of ecological momentary assessments (EMAs), which allow us to measure an individuals’ experiences in their natural environments, offering valuable data to inform just-in-time adaptive interventions (JITAIs). Machine learning can further enhance JITAIs by forecasting stress-related emotional states, enabling proactive intervention delivery to prevent heightened stress. In this study, we focus on forecasting stress utilizing data from a large mental health project. During this project, EMA data was collected from different vulnerable target groups across Europe, including youth, older adults, migrants, and individuals with low socioeconomic status. We formulated the forecasting task as a binary classification problem: predicting either transitions from normal to elevated stress or the stability of normal stress, based on a person's stress distribution. This approach simplifies the task, supports personalized predictions, and enables actionable insights, as predicting elevated stress can directly trigger support. Our results demonstrate that machine learning models are capable of forecasting stress transitions (ROC-AUC = 0.70 vs. 0.50 for a random classifier), although predicting transitions to elevated stress proved more challenging than identifying stable normal stress. Models trained on combined data from all populations performed comparable to those trained on individual populations. Furthermore, cross-country evaluations indicated that population-specific models generalized well across most populations. © 2026 The Authors