Article
Automation & Control Systems
SenPeng Chen, Jia Wu, XiYuan Liu
Summary: The paper proposes a novel method EMORL for hyperparameter optimization based on multi-objective reinforcement learning, which successfully addresses some limitations in traditional hyperparameter optimization. By combining accuracy and latency as a multi-objective reward, the policy update is effectively guided, resulting in improved optimization efficiency.
ENGINEERING APPLICATIONS OF ARTIFICIAL INTELLIGENCE
(2021)
Article
Computer Science, Artificial Intelligence
Chunnan Wang, Hongzhi Wang, Chang Zhou, Hanxiao Chen
Summary: Machine learning models are highly sensitive to hyperparameters and their evaluations can be costly. The ExperienceThinking algorithm proposed in this paper intelligently optimizes hyperparameter settings based on known evaluation information, effectively improving the performance of machine learning models within limited budgets.
KNOWLEDGE-BASED SYSTEMS
(2021)
Article
Automation & Control Systems
Matthias Feurer, Katharina Eggensperger, Stefan Falkner, Marius Lindauer, Frank Hutter
Summary: This paper introduces the background and development of AutoML, and proposes new approaches. The improvements of the proposed methods are validated through experimental research.
JOURNAL OF MACHINE LEARNING RESEARCH
(2022)
Article
Engineering, Multidisciplinary
Zeou Hu, Kiarash Shaloudegi, Guojun Zhang, Yaoliang Yu
Summary: In this work, federated learning is formulated as multi-objective optimization and a new algorithm called FedMGDA+ is proposed, which guarantees fairness and robustness while maintaining individual performance for participating users.
IEEE TRANSACTIONS ON NETWORK SCIENCE AND ENGINEERING
(2022)
Article
Engineering, Electrical & Electronic
Hendrik Wohrle, Felix Schneider, Fabian Schlenke, Denis Lebold, Mariela De Lucas Alvarez, Frank Kirchner, Michael Karagounis
Summary: In this paper, a methodology is proposed to co-optimize application-specific neural network accelerators for accuracy and energy efficiency per inference. The architecture of the neural network is co-optimized with the ASIC implementation of the accelerator to provide reliable estimates of energy efficiency. The method is demonstrated on an application-specific neural network accelerator for atrial fibrillation detection in electrocardiograms using 22FDX/FDSOI technology. The highly parameterizable neural network accelerator can map different neural networks with varying architectural properties to a synthesizable register transfer level representation. Different hyperparameter optimization methods, including Bayesian Optimization, are used and evaluated to find the optimal neural network architecture and physical implementation parameters.
IEEE TRANSACTIONS ON CIRCUITS AND SYSTEMS I-REGULAR PAPERS
(2023)
Article
Materials Science, Ceramics
Kensaku Nakamura, Naoya Otani, Tetsuya Koike
Summary: The study focuses on the trade-off between target properties of optical glass, using the ParEGO algorithm to find suitable glass compositions and optimize research components. The results show that ParEGO can effectively find compositions with low target property values, and normalization is necessary for search performance.
CERAMICS INTERNATIONAL
(2021)
Editorial Material
Computer Science, Artificial Intelligence
Xingyi Zhang, Ran Cheng, Liang Feng, Yaochu Jin
Summary: Optimization and learning are two main paradigms of artificial intelligence, frequently enhanced by each other in addressing complex real-world problems. Evolutionary multi-objective optimization algorithms are widely used but face challenges in solving complex problems. Machine learning techniques have been applied to enhance these algorithms.
IEEE COMPUTATIONAL INTELLIGENCE MAGAZINE
(2023)
Article
Multidisciplinary Sciences
Roa'a Mohammedqasem, Hayder Mohammedqasim, Sardar Asad Ali Biabani, Oguz Ata, Mohammad N. Alomary, Mazen Almehmadi, Ahad Amer Alsairi, Mohammad Azam Ansari
Summary: A deadly virus, COVID-19, originated in China and rapidly spread throughout the country. A new framework based on deep-learning optimization models was developed to handle medical datasets with high missing values, resulting in improved accuracy and efficiency. The experimental results demonstrated the effectiveness of the approach in classifying medical data sets and its potential applications in various fields.
JOURNAL OF KING SAUD UNIVERSITY SCIENCE
(2023)
Article
Engineering, Mechanical
Bin Hu, Zhaojie Wang, Chun Du, Wuyou Zou, Weidong Wu, Jianlin Tang, Jianping Ai, Huamin Zhou, Rong Chen, Bin Shan
Summary: In this study, optimization of TPMS structure for composite titania ceramic is accelerated using the multi-objective optimization algorithm guided finite element method simulation. The key factors for force reaction in TPMS structures are found to be thickness (P-t) and array number (P-a), while P-a mainly determines the pressure drop. The best performing TPMS structure with parameter combination (P-t=0.28, P-c=-0.49, P-a=3.5) yields suitable modulus and permeability required for practical bone filling application.
INTERNATIONAL JOURNAL OF MECHANICAL SCIENCES
(2023)
Article
Computer Science, Artificial Intelligence
Jia Wu, Xiyuan Liu, Senpeng Chen
Summary: In this paper, a context-based Meta-Reinforcement Learning approach combined with task-aware representation is proposed to address the challenges of data-inefficiency and limited generalization in hyperparameter optimization. The approach disentangles task inference and control, improves meta-training efficiency, and encourages smarter exploration. Experimental results demonstrate that the method can efficiently search for optimal hyperparameter configurations with limited computational cost.
KNOWLEDGE-BASED SYSTEMS
(2023)
Article
Computer Science, Artificial Intelligence
Yannik Zeitrag, Jose Rui Figueira, Nuno Horta, Rui Neves
Summary: This article introduces a new approach of reducing the computational costs of simulation-based fitness evaluation by utilizing surrogate models, and demonstrates the effectiveness and performance advantages of this method in solving dynamic job shop scheduling problems.
EXPERT SYSTEMS WITH APPLICATIONS
(2022)
Article
Engineering, Civil
Zhi Zheng, Shunyu Yao, Genghui Li, Linxi Han, Zhenkun Wang
Summary: Route planning is a key technology in intelligent transportation systems, and single-objective problems have been successfully solved using data-driven machine learning heuristics. However, practical route planning scenarios often involve multiple conflicting objectives. Existing research has proposed learning construction methods for multi-objective route planning (MORP), but learning improvement methods are lacking. To address this gap, this paper presents a learning improvement MORP method, Pareto Improver (PI), which utilizes a single deep reinforcement learning model to approximate the Pareto front. Experimental results demonstrate that PI outperforms other state-of-the-art methods.
IEEE TRANSACTIONS ON INTELLIGENT TRANSPORTATION SYSTEMS
(2023)
Article
Computer Science, Artificial Intelligence
Charles Moussa, Yash J. Patel, Vedran Dunjko, Thomas Baeck, Jan N. van Rijn
Summary: Researchers apply the functional ANOVA framework to analyze the importance of quantum machine learning hyperparameters in quantum neural network architectures. They find that the learning rate is the most important hyperparameter across all datasets, while the choice of entangling gates is the least important. They also develop data-driven priors based on previous dataset performance to guide hyperparameter optimization. This work introduces new methodologies for quantum model selection in practice.
Article
Computer Science, Artificial Intelligence
Xiyuan Liu, Jia Wu, Senpeng Chen
Summary: In this paper, a context-based meta-reinforcement learning approach is proposed to address the data-inefficiency problem in Hyperparameter Optimization (HPO). The approach utilizes an agent to sequentially select hyperparameters to maximize the expected accuracy of the machine learning algorithm. Experimental results demonstrate that the proposed approach outperforms other state-of-the-art optimization methods in terms of test set accuracy and runtime performance.
Article
Computer Science, Interdisciplinary Applications
Fabian Waschkowski, Yaomin Zhao, Richard Sandberg, Joseph Klewicki
Summary: This paper introduces two novel concepts in data-driven turbulence modeling, which enable the development of multiple closure models and training towards multiple objectives simultaneously. These concepts extend the evolutionary framework and apply a multi-objective optimization algorithm to achieve the coupling training of closure models and the balance of multiple training objectives. The results in benchmark cases show significant improvements, highlighting the importance of these concepts in achieving generalized data-driven turbulence models.
JOURNAL OF COMPUTATIONAL PHYSICS
(2022)
Article
Economics
Wen Shi, Kaijun Leng, Inneke Van Nieuwenhuyse, Yucui Liu, Xiaohong Chen
TRANSPORTATION RESEARCH PART A-POLICY AND PRACTICE
(2020)
Review
Computer Science, Interdisciplinary Applications
Sebastian Rojas-Gonzalez, Inneke Van Nieuwenhuyse
COMPUTERS & OPERATIONS RESEARCH
(2020)
Article
Management
Sebastian Rojas Gonzalez, Hamed Jalali, Inneke Van Nieuwenhuyse
EUROPEAN JOURNAL OF OPERATIONAL RESEARCH
(2020)
Article
Nursing
Chih-Hsuan Huang, Hsin-Hung Wu, Yii-Ching Lee, Inneke Van Nieuwenhuyse, Meng-Chen Lin, Cheng-Feng Wu
JOURNAL OF PEDIATRIC NURSING-NURSING CARE OF CHILDREN & FAMILIES
(2020)
Article
Computer Science, Interdisciplinary Applications
Nasrulloh Loka, Ivo Couckuyt, Federico Garbuglia, Domenico Spina, Inneke Van Nieuwenhuyse, Tom Dhaene
Summary: Multi-objective optimization of complex engineering systems is a challenging problem. Bayesian optimization is a popular technique to tackle this problem. We develop an approach that can handle a mix of expensive and cheap objective functions, offering lower complexity and superior performance in cases where the cheap objective function is difficult to approximate.
ENGINEERING WITH COMPUTERS
(2023)
Article
Engineering, Industrial
Hamed Jalali, Maud Van den Broeke, Inneke Van Nieuwenhuyse
Summary: The interaction between product platform and product portfolio decisions is crucial for a company's competitive advantage, but not well understood. Operational parameters and marketing parameters jointly impact the optimal product portfolio and platform design, and marketing parameters do not always affect the optimal product development strategy.
INTERNATIONAL JOURNAL OF PRODUCTION ECONOMICS
(2022)
Article
Green & Sustainable Science & Technology
Agnieszka Jastrzebska, Alejandro Morales Hernandez, Gonzalo Napoles, Yamisleydi Salgueiro, Koen Vanhoof
Summary: This paper proposes two new approaches for analyzing wind turbine health using time series processing and fuzzy sets. The methods aggregate and summarize raw data based on abstract concepts and observe changes in concepts to infer turbine health. Experimental results showed that turbines with IDs T07 and T06 degraded the most under relatively low atmospheric temperature and relatively high wind speed conditions.
Article
Computer Science, Artificial Intelligence
Alejandro Morales-Hernandez, Gonzalo Napoles, Agnieszka Jastrzebska, Yamisleydi Salgueiro, Koen Vanhoof
Summary: In this paper, a method based on Long Short-term Cognitive Networks (LSTCNs) is proposed for windmill time series forecasting. Compared to traditional Recurrent Neural Networks (RNNs), this method has faster speed and lower forecasting errors.
EXPERT SYSTEMS WITH APPLICATIONS
(2022)
Retraction
Computer Science, Information Systems
Kaijun Leng, Linbo Jin, Wen Shi, Inneke Van Nieuwenhuyse
CLUSTER COMPUTING-THE JOURNAL OF NETWORKS SOFTWARE TOOLS AND APPLICATIONS
(2022)
Proceedings Paper
Computer Science, Artificial Intelligence
Alejandro Morales-Hernandez, Inneke Van Nieuwenhuyse, Gonzalo Napoles
Summary: The performance of Machine Learning algorithms is influenced by the choice of hyperparameters. However, finding the optimal hyperparameters is challenging due to the expensive training and evaluation process. This paper proposes a method that combines Tree-structured Parzen Estimators (TPE) sampling strategy with Gaussian Process Regression (GPR) to optimize hyperparameters with uncertainty, leading to improved results compared to existing methods.
OPTIMIZATION AND LEARNING, OLA 2022
(2022)
Article
Business
Peng Xia, Zhixue Liu, Weijiao Wang, Wen Shi, Inneke Van Nieuwenhuyse
Summary: This study examines the factors driving vehicle recalls in the Chinese automobile industry and finds that firm innovation and negative electronic word-of-mouth play a significant role. The ownership structure of the firms also has an impact on the recall volume.
IEEE TRANSACTIONS ON ENGINEERING MANAGEMENT
(2022)
Proceedings Paper
Computer Science, Theory & Methods
Sebastian Rojas-Gonzalez, Juergen Branke, Inneke Van Nieuwehuyse
2019 WINTER SIMULATION CONFERENCE (WSC)
(2019)
