Accelerating Scientific Discovery with Machine Learning: From Methods to Applications

对研究背景的简要描述，包括潜在的影响：机器学习最近使从蛋白质折叠到核融合的科学取得了重大突破。然而，总的来说，在科学中应用数据驱动的方法仍然存在挑战，如有限的标记数据、多模态和约束模型以遵守已知的科学规律等。在本研究中，我们探索了将领域知识集成到机器学习模型中的方法，并相反，使用机器学习技术来帮助科学家解决问题-重点是人类-人工智能协作增强科学过程。通过这一点，我们希望使机器学习对领域专家来说更容易获得和相关，并加速科学发现。目的和目标：1.改进将科学归纳偏差和领域知识直接融入神经网络结构和训练的方法。应用机器学习技术来帮助解决特定领域的科学问题。研究方法的新颖性：我们将在科学领域的专业知识、物理模型、表征学习和人类-人工智能协作之间形成跨学科联系。与EPSRC的战略和研究领域保持一致：这项研究与EPSRC的许多研究领域保持一致，包括：人工智能技术、人机交互、信息系统、工程、卫生和政府的人工智能和数据科学(ASG)，以及统计和应用概率任何参与的公司或合作者：没有。

Brief description of the context of the research including potential impact:Machine learning has recently enabled significant breakthroughs in the sciences - from protein-folding to nuclear fusion. However in general, there remains challenges when applying data-driven methods in the sciences such as limited labelled data, multi-modality and constraining models to obey known scientific laws etc. In this research, we explore methods to integrate domain knowledge into machine learning models, and conversely, use machine learning techniques to help scientists solve problems - with a focus on human-AI collaboration augmenting the scientific process. Through this we hope to make machine learning more accessible and relevant to domain experts and accelerate scientific discovery.Aims and Objectives:1. Improve methods for incorporating scientific inductive biases and domain knowledge directly into neural network architectures and training.2. Apply machine learning techniques to help solve domain-specific scientific problems.Novelty of the research methodology:We will form cross-disciplinary links between scientific domain expertise, physical modelling, representation learning and human-AI collaboration.Alignment to EPSRC's strategies and research areas:This research is aligned to numerous EPSRC research areas including: Artificial Intelligence Technologies, Human-computer interaction, Information systems, AI and Data Science for Engineering, Health and Government (ASG), and Statistics and applied probabilityAny companies or collaborators involved:None.