Searcharxiv⌕ Search

arXiv subjects

Cameron Ryan

Publications and source records attributed to Cameron Ryan.

2 recordsLinked to original sources

High Pressure Material Phase Identification with Gaussian Kernel Neural Networks

Phase identification is a computational task involving the classification of material structure using the positions of atoms. Using data from Molecular Dynamics (MD) simulations, our task is to identify the material phase of a set of atoms, based on how the simulations predict their positioning relative to each other. By leveraging existing research on mechanistic neural network theory, particularly Neural Tangent Kernels (NTK), we develop a new and efficient method for classifying simulated material samples. Our method involves defining a set of "atom detectors" that are given positions around the origin in cartesian coordinates. The sum of Gaussians centered at each atom is calculated at each atom detector, and is used as the input to a neural network. We present theoretical and experimental evidence to show that this method is highly effective and efficient at the task. Our method obtains better or similar accuracy scores to commonly used methods while having a major computational speedup.

cond-mat.mtrl-sci↗

Unsupervised Latent Space Alignment with Hyperspherical Geodesic Matching

Independently trained neural networks tend to encode the same data with similar latent geometries. These latent geometries are not directly compatible, yet they can be nearly the same up to some class of transformations. While there exists many methods for alignment between different latent spaces, it is typically done using a set of shared sample correspondences, known as anchors. This leaves a fundamental question: are the geometric signatures of different latent spaces representing similar data sufficient to recover an alignment between them? To that end, we introduce HGA (Hyperspherical Gaussian Alignment), a method that directly optimizes a transformation between two latent spaces by maximizing a geometric measure of "fit" between them. Since it is driven by the geometry of the latent spaces rather than paired data, HGA can operate in both an unsupervised and weakly supervised regime. On tasks such as model stitching or multilingual word embedding correspondence recovery, HGA manages to match supervised results with minimal or no supervision.

cs.LG↗