An active machine learning approach for optimal design of magnesium alloys using Bayesian optimisation.

In the pursuit of magnesium (Mg) alloys with targeted mechanical properties, a multi-objective Bayesian optimisation workflow is presented to enable optimal Mg-alloy design. A probabilistic Gaussian process regressor model was trained through an active learning loop, while balancing the exploration and exploitation trade-off via an acquisition function of the upper confidence bound. New candidate alloys suggested by the optimiser within each iteration were appended to the training data, and the performance of this sequential strategy was validated via a regret analysis. Using the proposed approach, the dependency of the prediction error on the training data was overcome by considering both the predictions and their associated uncertainties. The method developed here, has been packaged into a web tool with a graphical user-interactive interface (GUI) that allows the proposed optimal Mg-alloy design strategy to be deployed.

Projected thickness reconstruction from a single defocused transmission electron microscope image of an amorphous object.

Single defocused transmission electron microscope phase contrast images are used to reconstruct the projected thickness map of a single-material object. The algorithm is non-iterative and stable, and we extend it to account for the presence of spherical aberration in the objective optics. The technique can reconstruct the projected thickness map of general single-material objects in the strong phase/weak amplitude regime. It is sensitive to any excursions in the projected thickness from the average, and ideal for examining voids and free volume accumulation in amorphous/glassy materials at the nanometer scale. The resolution of the technique depends on the choice of defocus and the thickness of the specimen. In a certain regime, we demonstrate that variations in the transverse projected thickness with a lateral diameter of ∼ 0.25 nm may be detected. We use our algorithm to quantitatively reconstruct the projected thickness of latex sphere test specimens from single defocused electron micrographs. We demonstrate that the reconstruction has a large tolerance for error in the input parameters. Simulations confirm that the technique is quantitative, and demonstrate that the origin of low-frequency artifacts is an instability due to noise. We show that the autocorrelation of the projected thickness map may be used to measure the size of open structures in the object using both simulation and latex sphere data.

Measuring the PSF from aperture images of arbitrary shape--an algorithm.

A new algorithm for determining the point spread function (PSF) of digital imaging systems is presented. The input is an image of an aperture whose shape need not be regular. The aperture shape is refined to an effective sub-pixel resolution and the PSF of the system is determined by de-convolution, assuming uniform illumination and a step function edge. The method has been tested on theoretical aperture images of varying shape and PSF, with and without noise. Depending on the degree of noise, a known PSF can be recovered to an accuracy of between 0.2 and 0.8%. Some typical results are given for a Gatan Image Filter with a 794 YAG multiscan camera on a Philips EM 430 transmission electron microscope at 200 and 300 kV. An example of a de-convoluted convergent beam electron diffraction pattern is included. The algorithm tolerates a small amount of de-focus.