We gratefully acknowledge support from
the Simons Foundation and member institutions.

Computational Geometry

New submissions

[ total of 2 entries: 1-2 ]
[ showing up to 2000 entries per page: fewer | more ]

New submissions for Thu, 13 May 21

[1]  arXiv:2105.05784 [pdf, other]
Title: Particle-Based Assembly Using Precise Global Control
Comments: 20 pages, 12 figures, full version of an extended abstract accepted for publication in the proceedings of the 17th Algorithms and Data Structures Symposium (WADS 2021)
Subjects: Computational Geometry (cs.CG); Data Structures and Algorithms (cs.DS)

In micro- and nano-scale systems, particles can be moved by using an external force like gravity or a magnetic field. In the presence of adhesive particles that can attach to each other, the challenge is to decide whether a shape is constructible. Previous work provides a class of shapes for which constructibility can be decided efficiently, when particles move maximally into the same direction on actuation.
In this paper, we consider a stronger model. On actuation, each particle moves one unit step into the given direction. We prove that deciding constructibility is NP-hard for three-dimensional shapes, and that a maximum constructible shape can be approximated. The same approximation algorithm applies for 2D. We further present linear-time algorithms to decide whether a tree-shape in 2D or 3D is constructible. If scaling is allowed, we show that the $c$-scaled copy of every non-degenerate polyomino is constructible, for every $c \geq 2$.

Cross-lists for Thu, 13 May 21

[2]  arXiv:2105.05275 (cross-list from cs.LG) [pdf, other]
Title: Hermitian Symmetric Spaces for Graph Embeddings
Comments: 13 pages, 1 figure. Accepted at NeurIPS 2020 workshop on Differential Geometry meets Deep Learning
Subjects: Machine Learning (cs.LG); Computational Geometry (cs.CG)

Learning faithful graph representations as sets of vertex embeddings has become a fundamental intermediary step in a wide range of machine learning applications. The quality of the embeddings is usually determined by how well the geometry of the target space matches the structure of the data. In this work we learn continuous representations of graphs in spaces of symmetric matrices over C. These spaces offer a rich geometry that simultaneously admits hyperbolic and Euclidean subspaces, and are amenable to analysis and explicit computations. We implement an efficient method to learn embeddings and compute distances, and develop the tools to operate with such spaces. The proposed models are able to automatically adapt to very dissimilar arrangements without any apriori estimates of graph features. On various datasets with very diverse structural properties and reconstruction measures our model ties the results of competitive baselines for geometrically pure graphs and outperforms them for graphs with mixed geometric features, showcasing the versatility of our approach.

[ total of 2 entries: 1-2 ]
[ showing up to 2000 entries per page: fewer | more ]

Disable MathJax (What is MathJax?)

Links to: arXiv, form interface, find, cs, recent, 2105, contact, help  (Access key information)