KI-Referenzmodell für Energie- und Ressourceneffizienz und dessen industrielle Anwendung (KIRA)

Publikationen im Rahmen des DMP KIRA

20232024

2024

[BFN+24] Bast, S., Fazlic, L. B., Naumann, S., & Guido Dartmann (2024). LLMs on the Edge: Quality, Latency, and Energy Efficiency. In INFORMATIK 2024. Bonn: Gesellschaft für Informatik e.V. Lecture Notes in Informatics (LNI). Accepted for publication.
Abstract: Generative Artificial Intelligence has become an integral part of many people’s lives. Large Language Models (LLMs) are gaining increasing popularity in science and society. While it is well known that training these models requires significant energy, inference also contributes to their total energy demand. Therefore, we analyze how to use them as sustainably as possible by investigating the efficiency of inference, especially on local hardware with limited computing power. We develop metrics for quantifying the efficiency of LLMs on the edge, focusing on the most influential factors quality, time, and energy. We compare the performance of three different state of the art generative models on the edge and assess the quality of the generated text, the time used for text creation and the energy demand down to the token level. The models achieve between 73,3% and 85,9% on the quality level, generate 1,83 to 3,51 tokens per second while consuming between 0,93 and 1,76 mWh of energy per token on a single-board computer without GPU-support. The findings of this study demonstrate that generative models can produce satisfactory outcomes on edge devices. However, a thorough efficiency evaluation is recommended before deploying them in production environments.
[CFG+24] Cetkin, B., Fazlic, L. B., Guldner, A., Naumann, S., & Dartmann, G. (2024). Towards Sustainable Machine Learning: Analyzing Energy-Efficient Algorithmic Strategies for Environmental Sensor Data. In INFORMATIK 2024. Bonn: Gesellschaft für Informatik e.V. Lecture Notes in Informatics (LNI). Accepted for publication.
Abstract: This study evaluates the energy efficiency of machine learning (ML) classification models across 49 test setups, each representing different conditions derived from a set of scenarios. Utilizing internet of things (IoT) technology with an ESP8266 microcontroller, we collected and analyzed environmental data including temperature, humidity, and CO2 levels from a simulated room environment. We measured energy consumption for data preprocessing, model training, and testing, alongside energy efficiency metrics that consider output, processing time, and F1 score. The study also performed correlation analyses to explore the relationship between energy consumption and performance metrics. Furthermore, it assessed the trade-offs between accuracy and energy efficiency by comparing an ensemble model to its constituent algorithms. The measurements, conducted according to the Green Software Measurement Model (GSMM), provide essential insights into selecting energy-efficient algorithms for a broad spectrum of IoT applications.
[CFU+24] Cetkin, B., Fazlic, L. B., Ueding, K., Machhamer, R., Guldner, A., Creutz, L., Naumann, S., & Dartmann, G. (2024). Spatial Impulse Response Analysis and Ensemble Learning for Efficient Precision Level Sensing. In: Discov Artif Intell 4, 63. https://doi.org/10.1007/s44163-024-00165-w
Abstract:In this paper, we propose an innovative method for determining the fill level of containers, such as trash cans, addressing a critical aspect of waste management. The method combines spatial impulse response analysis with machine learning (ML) techniques, offering a unique and effective approach for sound-based classification that can be extended to various domains beyond waste management. By employing a buzzer-generated sine sweep signal, we create a distinctive signature specific to the fill level of the waste container. This signature, once accurately decoded, is then interpreted by a specially developed ensemble learning algorithm. Our approach achieves a classification accuracy of over 90% when implemented locally on a development board, optimizing operational efficiencies and eliminating the need to delegate complex classification tasks to external entities. Using low-cost and energy-efficient hardware components, our method offers a cost-effective approach that contributes to sustainable and efficient waste management practices, providing a reliable and locally deployable solution.
[FCG+24] Fazlic, L. B., Cetkin, B., Guldner, A., Dziubany, M., Heinen, J., Naumann, S., & Dartmann, G. (2024). Enhancing Energy Efficiency in AI: A Multi-Faceted Analysis across Time Series, Semantic AI and Deep Learning Domains. Accepted at EnviroInfo 2024: 38th International Conference for Environmental Informatics. Cairo, Egypt, November 12-14, 2024.
Abstract: This research investigates strategies to enhance the energy efficiency of artificial intelligence (AI) algorithms, focusing on three pivotal domains: time series analysis, semantic AI, and deep learning (DL). Through a comprehensive examination of variables such as data size and the impact of hyper-parameter adjustments, the study aims to uncover nuanced insights into the relationship between algorithmic performance and energy consumption. By exploring the unique challenges and opportunities within each use case, this research provides valuable guidance for practitioners seeking to optimize energy efficiency in AI applications. The findings contribute to the ongoing discourse on sustainable AI development, offering practical overview to balance computational power with environmental considerations.
[GBC+24] Guldner, A., Bender, R., Calero, C., Fernando, G. S., Funke, M., Gröger, J., Hilty, L. M., Hörnschemeyer, J., Hoffmann, G.-D., Junger, D., Kennes, T., Kreten, S., Lago, P., Mai, F., Malavolta, I., Murach, J., Obergöker, K., Schmidt, B., Tarara, A., ... Naumann, S. (2024). Development and evaluation of a reference measurement model for assessing the resource and energy efficiency of software products and components—Green Software Measurement Model (GSMM). In Future Generation Computer Systems (Vol. 155, pp. 402–418). Elsevier BV. doi.org/10.1016/j.future.2024.01.033
Abstract: In the past decade, research on measuring and assessing the environmental impact of software has gained significant momentum in science and industry. However, due to the large number of research groups, measurement setups, procedure models, tools, and general novelty of the research area, a comprehensive research framework has yet to be created. The literature documents several approaches from researchers and practitioners who have developed individual methods and models, along with more general ideas like the integration of software sustainability in the context of the UN Sustainable Development Goals, or science communication approaches to make the resource cost of software transparent to society. However, a reference measurement model for the energy and resource consumption of software is still missing. In this article, we jointly develop the Green Software Measurement Model (GSMM), in which we bring together the core ideas of the measurement models, setups, and methods of over 10 research groups in four countries who have done pioneering work in assessing the environmental impact of software. We briefly describe the different methods and models used by these research groups, derive the components of the GSMM from them, and then we discuss and evaluate the resulting reference model. By categorizing the existing measurement models and procedures and by providing guidelines for assimilating and tailoring existing methods, we expect this work to aid new researchers and practitioners who want to conduct measurements for their individual use cases.
[MBG+24] Machhamer, R., Begic Fazlic, L., Guven, E., Junk, D., Karabulut Kurt, G., Naumann, S., Didas, S., Gollmer, K.-U., Bergmann, R., Timm, I. J., & Dartmann, G. (2024). Likelihood-Based Sensor Calibration Using Affine Transformation. In IEEE Sensors Journal (Vol. 24, Issue 3, pp. 3672–3680). Institute of Electrical and Electronics Engineers (IEEE). https://doi.org/10.1109/jsen.2023.3341503
Abstract: An important task in the field of sensor technology is the efficient implementation of adaptation procedures of measurements from one sensor to another sensor of identical design. One idea is to use the estimation of an affine transformation (AT) between different systems, which can be improved by the knowledge of experts. This article presents an improved solution from Glacier Research that was published back in 1973. The results demonstrate the adaptability of this solution for various applications, including software calibration of sensors, implementation of expert-based adaptation, and paving the way for future advancements such as distributed learning methods. One idea here is to use the knowledge of experts for estimating an AT between different systems. We evaluate our research with simulations and also with real measured data of a multisensor board with eight identical sensors. Both dataset and evaluation script are provided for download. The results show an improvement for both the simulation and the experiments with real data.
[MFD+24] Morgen, M., Fazlic, L. B., & Dartmann, G. (2024). Connect, Understand and Learn: Dynamic Knowledge Graph Transforms Learning. In 2024 47th MIPRO ICT and Electronics Convention (MIPRO). 2024 47th MIPRO ICT and Electronics Convention (MIPRO). IEEE. https://doi.org/10.1109/mipro60963.2024.10569675
Abstract: The automation of knowledge graphs is a challenge if only small training data sets are available for the corresponding learning methods. The approach presented in this paper can work with small training data sets and enables the solution of tasks with previously hidden syntactic structures. In this research, a new conceptual algorithm for learning and updating knowledge graphs is proposed. We combined a powerful NLP approach with statistical methods to build a word frequency-based corpus for various question answering problems. Then, we used specific similarity measures to find the best answer for the given problem. For this purpose, a vector model is used and weights are calculated for the association between terms and problems. In the last phase, we created a continuous learning model with a dynamic knowledge graph that can be updated with new tasks and predict answers to upcoming problems. The knowledge graph is updated with new information when the pattern of a problem is unknown and therefore not found. The implementation of the algorithm is validated using various openly available data sets from the field of user support in business and medicine. The proposed method supports an incremental learning approach and real-time implementation.

2023

[WGB+23] Weber, S., Guldner, A., Begic Fazlic, L., Dartmann, G., & Naumann, S. (2023). Sustainability in Artificial Intelligence - Towards a Green AI Reference Model. In Designing Futures: Zukünfte gestalten. Bonn: Gesellschaft für Informatik e.V. Lecture Notes in Informatics (LNI). PISSN: 1617-5468. ISBN: 978-3-88579-731-9. pp. 1503-1514. Ökologische Nachhaltigkeit - KIU-2026. Berlin. 26.-29. September 2023 https://doi.org/10.18420/INF2023_154
Abstract: The interest in Green Artificial Intelligence (AI) is growing as AI research is increasingly focusing on and taking into account environmental sustainability. This paper aims to clarify and emphasize the distinction between terms like sustainable AI, Green AI, Green by AI, and Green in AI, highlighting their importance in the context of environmentally responsible AI practices. We find that existing Green Software reference models are insufficient for meeting the unique requirements of Green AI. Thus, we argue that a tailored Green AI reference model is needed to guide and promote environmentally responsible practices in the field of AI, addressing the special considerations associated with Green AI.
back-to-top nach oben