この続きは、会員限定コンテンツです。
会員登録をすると、全文をご覧いただけます。
Bayesian Optimization (BO) is effective for costly evaluation problems, but traditional BO performance degrades under extremely limited budgets due to insufficient sampling, poor exploration-exploitation balance, and the curse of dimensionality. To address this "limited-budget BO" challenge, research explores "Search Space Refinement" (focusing on promising regions) and "Look-ahead (Non-myopic)" approaches (considering future steps when selecting points). While potentially improving evaluation efficiency, these methods introduce new challenges like increased computational cost, approximation errors, tuning needs, and scalability issues. Practical application, especially in Materials Informatics (MI), requires careful method selection based on problem specifics and consideration of these trade-offs.
Sultana Sumaiya Saima
MI-6 Ltd.Machine Learning Research
She holds a Bachelor's degree in Mechanical Engineering from the Bangladesh University of Engineering and Technology (BUET). Following her undergraduate studies, she transitioned into the field of artificial intelligence, with a particular focus on the industrial applications of machine learning and deep learning techniques.
Professionally, she has worked as an AI Engineer, where she contributed to the development and standardisation of quality assurance pipelines for AI systems, ensuring robust performance from initial development through to production deployment.
In her current role as a Machine Learning Researcher, her work centres on the research and development of advanced optimisation algorithms tailored for black-box systems within the field of materials informatics. Her research aims to enhance data-driven methodologies to accelerate materials discovery and design.
人気の記事
お役立ち資料
Loading...
