Bayesian Active Learning for Scanning Probe Microscopy: From Gaussian Processes to Hypothesis Learning

Abstract

Recent progress in machine learning methods and the emerging availability of programmable interfaces for scanning probe microscopes (SPMs) have propelled automated and autonomous microscopies to the forefront of attention of the scientific community. However, enabling automated microscopy requires the development of task-specific machine learning methods, understanding the interplay between physics discovery and machine learning, and fully defined discovery workflows. This, in turn, requires balancing the physical intuition and prior knowledge of the domain scientist with rewards that define experimental goals and machine learning algorithms that can translate these to specific experimental protocols. Here, we discuss the basic principles of Bayesian active learning and illustrate its applications for SPM. We progress from the Gaussian process as a simple data-driven method and Bayesian inference for physical models as an extension of physics-based functional fits to more complex deep kernel learning methods, structured Gaussian processes, and hypothesis learning. These frameworks allow for the use of prior data, the discovery of specific functionalities as encoded in spectral data, and exploration of physical laws manifesting during the experiment. The discussed framework can be universally applied to all techniques combining imaging and spectroscopy, SPM methods, nanoindentation, electron microscopy and spectroscopy, and chemical imaging methods and can be particularly impactful for destructive or irreversible measurements.

Adaptive sampling for accelerating neutron diffraction-based strain mapping *

“…We implemented our system in an experimental set up and demonstrated that the strain mapping task can be performed in nearly 2.4 h instead of the typical 6 h for a specific study on an AM steel part. Our work builds on a larger trend of moving towards autonomous scanning probe imaging systems [18,25] in cases where the time to measure is significantly higher than the sum of the time to re-position the object and computation of the next informative measurement to make. In such systems, there can be an overall saving in experimental measurement time while providing a progressively improved image resolution by employing a powerful algorithm that can infer the underlying structure from a sparse set of measurement points.…”

“…where X prev is the set of all the previous locations on the grid at which a measurement has been made. This type of approach is called BO [25] because we effectively compute the posterior distribution parameters based on some underlying prior model to infer the next point(s) to measure. A simple acquisition function is one that simply uses the variance estimates ('uncertainty') at each position from equation ( 5) and chooses to measure the next point at the position corresponding to the highest variance.…”

Deep kernel methods learn better: from cards to process optimization

“…DKL poses to be one of the potential solutions to the problems that are discussed so far. DKL is a method that combines the power of deep neural networks with the flexibility of Gaussian Processes (GPs) to model and make predictions on complex target functions in high-dimensional spaces [59][60][61][62][63]. The ability to efficiently explore the high-dimensional input space and learn the relationship between the inputs and target function, makes it a powerful tool for optimization and prediction tasks.…”

A Processing and Analytics System for Microscopy Data Workflows: The Pycroscopy Ecosystem of Packages

“…Moreover, much of modern microscopy has advanced with applications of deep learning [16] and other machine learning methods, [17] assisting in everything from data cleaning, [18] to improved functional fits, [19] image segmentation, [20] and recently, to autonomous microscopy operations. [21,22] Therefore, datasets should be able to be used in distributed systems or supercomputer centers. As such, with the recent advancement in automated platforms, standardizing the data model and processing pipelines is an imperative for maximizing the utility of microscopy, and for reducing the significant bottlenecks in the analysis, storage, processing, and retrieval that are currently within traditional experimental paradigms.…”