Search

Machine learning is rapidly becoming an integral part of experimental physical discovery via automated and high-throughput synthesis, and active experiments in scattering and electron/probe microscopy. This, in turn, necessitates the development of active learning methods capable of exploring relevant parameter spaces with the smallest number of steps. Here, an active learning approach based on conavigation of the hypothesis and experimental spaces is introduced. This is realized by combining the structured Gaussian processes containing probabilistic models of the possible system's behaviors (hypotheses) with reinforcement learning policy refinement (discovery). This approach closely resembles classical human-driven physical discovery, when several alternative hypotheses realized via models with adjustable parameters are tested during an experiment. This approach is demonstrated for exploring concentration-induced phase transitions in combinatorial libraries of Sm-doped BiFeO 3 using piezoresponse force microscopy, but it is straightforward to extend it to higher-dimensional parameter spaces and more complex physical problems once the experimental workflow and hypothesis generation are available., (© 2022 Wiley-VCH GmbH.)

Methane is a kind of harmful gas produced in the process of coal mining. Because of its flammable and explosive characteristics, it poses a great threat to safety during coal mining. In practice, accurate and real-time gas concentration forecasting is becoming an essential issue for reducing methane risks and accidents. Given the complex spatial correlation and temporal variation of sensor data in a coal mine monitoring system, deep learning algorithms have been widely applied due to their revolutionary feature representation capability. However, existing deep learning models utilize recurrent neural networks, which can barely provide satisfactory accuracy due to their ignorance of realistic working conditions of a coal face or an insufficiency in capturing representative spatio-temporal patterns. In this paper, we propose an attention-based spatio-temporal encoder–decoder network approach, named the ASTED model, for methane concentration forecasting. The ASTED model is built based on the integration of the spatial, temporal and environmental information. Specifically, the multi-attention mechanism is used to learn the dynamic spatio-temporal dependencies, and the feature fusion module is used to incorporate the data from different mine sensors. Finally, we employ the LSTM-based encoder–decoder model to generate the final prediction results. Experiments demonstrate that the ASTED model can obtain the dynamic spatio-temporal correlation from multiple sensor readings and achieve the best performance compared with various state-of-the-art solutions. [ABSTRACT FROM AUTHOR]