Dec 2026· IEEE transactions on computers· Vol 75, pp. 4336-4348· 0 citations· 43 references
TL;DR
DGRL designs a novel mechanism to realize distributed inference and training of GNNs, making it possible to generate site-wise deployment policies and drastically improve scalability for coupled decision-making, and exhibits superior scalability compared to federated learning.
Abstract
Microservice (MS) architecture breaks down an application into lightweight service modules, making it an ideal choice for delivering complex applications in resource-constrained mobile edge computing (MEC) networks. This paper studies an Edge Micro-Service (EdgeMS) platform with a specific focus on EdgeMS deployment. EdgeMS deployment refers to installing MS modules at geo-distributed edge servers and enables cross-site function calls to accomplish the full functionalities. This problem is particularly challenging due to the implicit impact of disunified platform configurations, the complicated interactions of MS modules over dynamic MEC networks, and the deployment coupling across edge servers. We propose a novel algorithm called Distributed Graph Reinforcement Learning (DGRL) to handle the above unique challenges. DGRL employs a reinforcement learning framework that learns an EdgeMS deployment policy by directly interacting with the EdgeMS platform, thereby avoiding inaccurate characterizations of various influencing factors under disunified platform configurations. DGRL embeds graph neural networks as its learning engine, enabling it to capture complicated interactions between MS modules over dynamic edge networks. Particularly, DGRL designs a novel mechanism to realize distributed inference and training of GNNs, making it possible to generate site-wise deployment policies and drastically improve scalability for coupled decision-making. We evaluate the performance of DGRL on a realistic MEC network topology, utilizing real-world MS applications, and compare it against state-of-the-art learning strategies in both centralized and distributed settings. The results demonstrate that DGRL reduces the data transmission delay and computing delay 24.58× and 10.19×, respectively, during learning, while maintaining performance comparable to centralized learning. Additionally, DGRL exhibits superior scalability compared to federated learning.
This publication proposes a definition and a classification of agile software development approaches and analyses ten software development methods that can be characterized as being "agile" against the defined criterion.
P. Abrahamsson, O. Salo, Jussi Ronkainen et al.· arXiv.org· 727 citations· ⚡54
The study shows that agile practices improve both informal and formal communication, but indicates that, in larger development situations involving multiple external stakeholders, a mismatch of adequate communication mechanisms can sometimes even hinder the communication.
M. Pikkarainen, Jukka Haikara, O. Salo et al.· Empirical Software Engineeri...· 401 citations· ⚡48
The results indicate that software engineering work practices are chosen opportunistically, adapted and configured to provide value under the constrains imposed by the startup context.
Nicolò Paternoster, Carmine Giardino, M. Unterkalmsteiner et al.· Information and Software Tec...· 394 citations· ⚡54
The perception of the impact of agile methods is predominantly positive, and several challenge areas were discovered, but based on this study, agile methods are here to stay.
M. Laanti, O. Salo, P. Abrahamsson· Information and Software Tec...· 260 citations· ⚡20
Related blog posts
Microsoft Research Blog· microsoft.comJul 13, 2026
Cryptographic code supports vital protections in modern computing systems. Learn how a new method helps verify code as developers write it while preserving speed and adaptability as it gets implemented and evolves. The post Verifying Rust cryptography in SymCrypt, from standards to code appeared first on Microsoft Research.
Machine-learning networks can learn without detailed instructions from a designer. Those that do may offer insight into how the brain reorganizes itself during learning.
MIT News · Artificial Intelligence· news.mit.eduOct 2, 2026