Distributed Algorithms for AI Accelerated Materials Discovery

在快速发展的化学发现领域，高通量实验数据与计算模拟结果的结合为加快新化合物的识别和了解其性质提供了一条很有前途的途径。这项研究引入了一个新的人工智能框架，利用多保真贝叶斯机器学习模型，包括高斯过程(GP)和贝叶斯神经网络(BNN)，以协调从不同的数据来源-实验室实验和计算机驱动的模拟-获得的见解。通过使用多输出贝叶斯方法，我们的方法唯一地捕获了不同物理和化学结果之间的相关性，以及不同保真度数据集中的固有偏差。通过它在预测新化学实体的性质方面的应用，从而为药物发现和材料科学的实验设计和决策过程提供信息，展示了该框架的实用性。我们的发现表明，预测的准确性和效率有了实质性的提高，这突显了多保真模型通过使实验数据和模拟数据能够更有效地融合而彻底改变化学发现的潜力。这项工作不仅促进了人工智能在化学中应用的方法学进步，而且为未来旨在优化计算模拟和经验实验之间的协同以加速发现的研究设定了新的基准

In the rapidly evolving field of chemical discovery, the integration of high-throughput experimental data with computational simulation results offers a promising avenue to accelerate the identification of novel compounds and understand their properties. This research introduces a novel artificial intelligence framework leveraging multi-Fidelity Bayesian machine learning models, including Gaussian Processes (GPs) and Bayesian Neural Networks (BNNs), to harmonize insights drawn from disparate data sources-laboratory experiments and computer-driven simulations. By employing Multi-Output Bayesian methods, our approach uniquely captures the correlations between different physical and chemical outcomes, as well as the inherent biases in datasets of varying fidelity. The framework's utility is demonstrated through its application in predicting the properties of new chemical entities, thereby informing experimental design and decision-making processes in drug discovery and materials science. Our findings indicate a substantial enhancement in prediction accuracy and efficiency, underscoring the potential of multi-fidelity models to revolutionize chemical discovery by enabling a more effective fusion of experimental and simulated data. This work not only contributes to the methodological advancements in the application of AI in chemistry but also sets a new benchmark for future research aiming to optimize the synergy between computational simulations and empirical experiments for accelerated discovery