Recent developments in experimental and simulation tools have converged with advances in raw computing resources, necessitating the development of new algorithms for analyzing the data generated. The ability to collect more digital data at faster speeds with multiple signals, length scales and viewing angles, does not lead to improved materials characterization or modeling without the ability to process and understand the data. This is becoming increasingly apparent as sensors are capturing data at a rate which cannot effectively be analyzed by a human. Indeed, when Materials Science is viewed as a "Big Data" problem, it becomes immediately apparent that our field presents problems unique to Materials Science and that, because of the highly complex and non-linear relationships known to exist, a simple correlation is usually dissatisfying. Robust analytical methods for extraction of quantitative physical information from raw data, and not just correlations, are needed. This type of analysis is inaccessible by traditional methods, either due to massive data volumes or, conversely, missing data.
From an experimental perspective, all measurements become inverse problems: the end result of the interaction of a beam with a material is all that is observed and the material structure/composition which gave produced the result needs to be found. Similarly, computational design of materials to meet property targets can be framed as an inverse problem, using physics based models as forward models in the inversion. Experiments, by their very nature, have limitations such as indirect measurements, inefficient detectors, restricted field of view, sample damage, and noise. This makes for ambiguous interpretation from a machine's point of view- i.e. the inversion is ill-posed and the results from traditional analysis methods give uninterpretable or physically unrealistic results.
This symposium will cover advances in methods for data analytics, for both experimentally and computationally generated data, specifically as they have been applied to materials science problems. This symposium will focus on current challenges in data analytics for materials science such as high throughput data generation, inverse methods, three and four dimensional data, multimodal data, multi-physics model data, and others.