Adaptive AI and Decision Intelligence for Sustainable Systems and Green IS

Permanent URI for this collectionhttps://hdl.handle.net/10125/112415

Browse

Recent Submissions

Now showing 1 - 5 of 5
  • Thumbnail Image
    Item type: Item ,
    A Scalable Distillation and Metric-Learning Pipeline for Adaptive Weed Classification on Edge Devices
    (2026-01-06) Allalen, Abderrezak; El-Gayar, Omar
    Smart agriculture increasingly relies on automated weed detection to reduce inputs and labor. Deploying deep learning on edge devices is difficult due to limited compute and evolving weed classes. We propose a pipeline that combines partial fine-tuning of an EfficientNet-B7 teacher, embedding level distillation into lightweight students (MobileNetV3, ShuffleNetV2, EfficientNet-B0), and semi-hard triplet metric learning. The system learns 2048D embeddings and is evaluated with N-way/K-shot episodes to mimic a few-label condition. Dynamic INT8 quantization enables CPU-only inference with minimal accuracy loss. The approach adapts rapidly to novel species with few labels while meeting real-time edge constraints, supporting sustainable herbicide management in practice.
  • Thumbnail Image
    Item type: Item ,
    Large Language Models Based In-Context Learning for Early Stage Building Life Cycle Assessment
    (2026-01-06) Jung, Hyeyun Eunice; Yu, Muran; Wang, Jie; Law, Kincho; Lepech, Michael D
    This research investigates applying Large Language Models (LLMs) using in-context learning for early stage building Life Cycle Assessment (LCA). Building LCA is challenging at the early stage since information is limited and few decisions have been made. Traditionally, building LCA follows a linear workflow in which architects transfer designs to engineers for calculation, resulting in a time-consuming, ineffective process due to the knowledge gap between disciplines. Our study presents three LLM methods: innate, domain, and ontology-based knowledge approaches. We demonstrate that the ontology-based approach, utilizing an iterative process, reduces the Root Mean Square Error (RMSE) by 41% and standard deviation by 76% compared to the innate knowledge method. Our findings highlight the potential of LLMs as AI agents to integrate expert knowledge in early design by systematically providing domain knowledge. In-context learning represents a novel approach to addressing limitations in the early building LCA, with potential applications across engineering fields.
  • Thumbnail Image
    Item type: Item ,
    Toward a Holistic Conceptualization of Green Coding: A Multi-Level Perspective on Sustainable Software Practices
    (2026-01-06) Gurung, Ram; Khan, Muhammad Asif; Siemon, Dominik; Porras, Jari
    The rising use of information and communication technology (ICT) is contributing to global greenhouse gas emissions (GHGEs), with projections indicating further growth. As ICT systems are software-driven, improving software energy efficiency through green coding has become more important. However, research on green coding remains disintegrated, focusing mainly on tools, green coding practices, implementation in specific phases of the software development life cycle (SDLC), and specific application domains. This paper addresses the conceptual gap by offering a multi-level conceptualization of green coding through the theoretical lens of responsible innovation, sociotechnical systems theory, organizational theory, and institutional theory. Our conceptual framework highlights how technical skills, sustainability mindsets, and ethical values influence the adoption of green coding, shaped by internal organizational dynamics and external institutional pressures. It also explores how green coding has varying impacts across application domains. We propose three propositions and offer a foundation for advancing sustainable software development.
  • Thumbnail Image
    Item type: Item ,
    How Stakeholders Perceive Generative AI in Sustainability Reports: Assistance or Interference?
    (2026-01-06) Brune, Niclas; Hinsberger, Björn
    Progressive improvements in Generative Artificial Intelligence (GenAI) are leading to an expansion of its application, including sustainability reporting. While this promises efficiency gains, little is known about how stakeholders perceive AI-generated reports. This study investigates the perception of external stakeholders of such reports with different levels of AI involvement, examined through an online experiment with 96 participants in Germany. Our findings show that human post-processing plays a crucial role: reports co-created with AI are perceived as equally credible as human-written ones, while fully automated reports are rated significantly lower. These results underline the relevance of human involvement for maintaining credibility in sensitive communication contexts. They also provide practical insights for communicating AI-assisted sustainability reporting, highlighting the psychological dynamics that shape stakeholder trust.
  • Thumbnail Image
    Item type: Item ,