Scholarlink

<p>Self-driving laboratories (SDLs) represent a cutting-edge convergence of machine learning with laboratory automation. SDLs operate in active learning loops, in which a machine learning algorithm plans experiments that are subsequently executed by increasingly automated (robotic) modules. Here we present our view on emerging SDLs for accelerated discovery and process optimization in heterogeneous catalysis. We argue against the paradigm of full automation and the goal of keeping the human out of the loop. Based on analysis of the involved workflows, we instead conclude that crucial advances will come from establishing fast proxy experiments and re-engineering existing apparatuses and measurement protocols. Industrially relevant use cases will also require humans to be kept in the loop for continuous decision-making. In turn, active learning algorithms will have to be advanced that can flexibly deal with corresponding adaptations of the design space and varying information content and noise in the acquired data.</p><p></p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Role of the human-in-the-loop in emerging self-driving laboratories for heterogeneous catalysis

  • 期刊论文

  • 发表日期：

    2025年1月29日

查看全文
  • Christoph Scheurer,
  • Karsten Reuter

摘要

Self-driving laboratories (SDLs) represent a cutting-edge convergence of machine learning with laboratory automation. SDLs operate in active learning loops, in which a machine learning algorithm plans experiments that are subsequently executed by increasingly automated (robotic) modules. Here we present our view on emerging SDLs for accelerated discovery and process optimization in heterogeneous catalysis. We argue against the paradigm of full automation and the goal of keeping the human out of the loop. Based on analysis of the involved workflows, we instead conclude that crucial advances will come from establishing fast proxy experiments and re-engineering existing apparatuses and measurement protocols. Industrially relevant use cases will also require humans to be kept in the loop for continuous decision-making. In turn, active learning algorithms will have to be advanced that can flexibly deal with corresponding adaptations of the design space and varying information content and noise in the acquired data.