An official website of the United States government
Many metric learning tasks, such as triplet learning, nearest neighbor retrieval, and visualization, are treated primarily as embedding tasks where the ultimate metric is some variant of the Euclidean distance (e.g., cosine or Mahalanobis), and the algorithm must learn to embed points into the pre-chosen space. The study of non-Euclidean geometries is often not explored, which we believe is due to a lack of tools for learning non-Euclidean measures of distance. Recent work has shown that Bregman divergences can be learned from data, opening a promising approach to learning asymmetric distances. We propose a new approach to learning arbitrary Bergman divergences in a differentiable manner via input convex neural networks and show that it overcomes significant limitations of previous works. We also demonstrate that our method more faithfully learns divergences over a set of both new and previously studied tasks, including asymmetric regression, ranking, and clustering. Our tests further extend to known asymmetric, but non-Bregman tasks, where our method still performs competitively despite misspecification, showing the general utility of our approach for asymmetric learning.
more »
« less
Crystallographic variant mapping using precession electron diffraction data
In this work, we developed three methods to map crystallographic variants of samples at the nanoscale by analyzing precession electron diffraction data using a high-temperature shape memory alloy and a VO2 thin film on sapphire as the model systems. The three methods are (I) a user-selecting-reference pattern approach, (II) an algorithm-selecting-reference-pattern approach, and (III) a k-means approach. In the first two approaches, Euclidean distance, Cosine, and Structural Similarity (SSIM) algorithms were assessed for the diffraction pattern similarity quantification. We demonstrated that the Euclidean distance and SSIM methods outperform the Cosine algorithm. We further revealed that the random noise in the diffraction data can dramatically affect similarity quantification. Denoising processes could improve the crystallographic mapping quality. With the three methods mentioned above, we were able to map the crystallographic variants in different materials systems, thus enabling fast variant number quantification and clear variant distribution visualization. The advantages and disadvantages of each approach are also discussed. We expect these methods to benefit researchers who work on martensitic materials, in which the variant information is critical to understand their properties and functionalities.
more »
« less
- Award ID(s):
- 2004752
- PAR ID:
- 10435200
- Date Published:
- Journal Name:
- Microstructures
- Volume:
- 3
- Issue:
- 4
- ISSN:
- 2770-2995
- Page Range / eLocation ID:
- 2023029
- Format(s):
- Medium: X
- Sponsoring Org:
- National Science Foundation
More Like this
-
-
Electron backscatter diffraction (EBSD) is a powerful tool for determining the orientations of near-surface grains in engineering materials. However, many ceramics present challenges for routine EBSD data collection and indexing due to small grain sizes, high crack densities, beam and charge sensitivities, low crystal symmetries, and pseudo-symmetric pattern variants. Micro-cracked monoclinic hafnia, tetragonal hafnon, and hafnia/hafnon composites exhibit all such features, and are used in the present work to show the efficacy of a novel workflow based on a direct detecting EBSD sensor and a state-of-the-art pattern indexing approach. At 5 and 10 keV primary beam energies (where beam-induced damage and surface charge accumulation are minimal), the direct electron detector produces superior diffraction patterns with 10x lower doses compared to a phosphor-coupled indirect detector. Further, pseudo-symmetric variant-related indexing errors from a Hough-based approach (which account for at least 4%-14% of map areas) are easily resolved by dictionary indexing. In short, the workflow unlocks fundamentally new opportunities to characterize materials historically unsuited for EBSD.more » « less
-
High resolution electron backscatter diffraction (HR-EBSD) is a technique used to map elastic strain, crystallographic orientation and dislocation density in a scanning electron microscope. This review covers the background and mathematics of this technique, contextualizing it within the broader landscape of EBSD techniques and other materials characterization methods. Several case studies are presented showing the application of HR-EBSD to the study of plasticity in metals, failure analysis in microelectronics and defect quantification in thin films. This is intended to be a comprehensive resource for researchers developing this technique as well as an introduction to those wishing to apply it.more » « less
-
Route reconstruction is an important application for Geographic Information Systems (GIS) that rely heavily upon GPS data and other location data from IoT devices. Many of these techniques rely on geometric methods involving the \frechet\ distance to compare curve similarity. The goal of reconstruction, or map matching, is to find the most similar path within a given graph to a given input curve, which is often approximate location data. This process can be approximated by sampling the curves and using the \dfd. Due to power and coverage constraints, the GPS data itself may be sparse causing improper constraints along the edges during the reconstruction if only the continuous \frechet\ distance is used. Here, we look at two variations of discrete map matching: one constraining the walk length and the other limiting the number of vertices visited in the graph. %, and the constraint that the walk may not self-intersect. We give an efficient algorithm to solve the question based on walk length showing it is in \textbf{P}. We prove the other problem is \npc\ and the minimization variant is \apx\ while also giving a parameterized algorithm to solve the problem.more » « less
-
null (Ed.)The 2-Wasserstein distance (or RMS distance) is a useful measure of similarity between probability distributions with exciting applications in machine learning. For discrete distributions, the problem of computing this distance can be expressed in terms of finding a minimum-cost perfect matching on a complete bipartite graph given by two multisets of points A, B ⊂ ℝ2, with |A| = |B| = n, where the ground distance between any two points is the squared Euclidean distance between them. Although there is a near-linear time relative ∊-approximation algorithm for the case where the ground distance is Euclidean (Sharathkumar and Agarwal, JACM 2020), all existing relative ∊-approximation algorithms for the RMS distance take Ω(n3/2) time. This is primarily because, unlike Euclidean distance, squared Euclidean distance is not a metric. In this paper, for the RMS distance, we present a new ∊-approximation algorithm that runs in O(n^5/4 poly{log n, 1/∊}) time. Our algorithm is inspired by a recent approach for finding a minimum-cost perfect matching in bipartite planar graphs (Asathulla et al, TALG 2020). Their algorithm depends heavily on the existence of sublinear sized vertex separators as well as shortest path data structures that require planarity. Surprisingly, we are able to design a similar algorithm for a complete geometric graph that is far from planar and does not have any vertex separators. Central components of our algorithm include a quadtree-based distance that approximates the squared Euclidean distance and a data structure that supports both Hungarian search and augmentation in sublinear time.more » « less
- Free Publicly Accessible Full Text
-
Accepted Manuscript1.0
- Journal Article:
- https://doi.org/10.20517/microstructures.2023.17
