-
A Practical Guide to Statistical Distances for Evaluating Generative Models in Science
Authors:
Sebastian Bischoff,
Alana Darcher,
Michael Deistler,
Richard Gao,
Franziska Gerken,
Manuel Gloeckler,
Lisa Haxel,
Jaivardhan Kapoor,
Janne K Lappalainen,
Jakob H Macke,
Guy Moss,
Matthijs Pals,
Felix Pei,
Rachel Rapp,
A Erdem Sağtekin,
Cornelius Schröder,
Auguste Schulz,
Zinovia Stefanidi,
Shoji Toyota,
Linda Ulmer,
Julius Vetter
Abstract:
Generative models are invaluable in many fields of science because of their ability to capture high-dimensional and complicated distributions, such as photo-realistic images, protein structures, and connectomes. How do we evaluate the samples these models generate? This work aims to provide an accessible entry point to understanding popular notions of statistical distances, requiring only foundati…
▽ More
Generative models are invaluable in many fields of science because of their ability to capture high-dimensional and complicated distributions, such as photo-realistic images, protein structures, and connectomes. How do we evaluate the samples these models generate? This work aims to provide an accessible entry point to understanding popular notions of statistical distances, requiring only foundational knowledge in mathematics and statistics. We focus on four commonly used notions of statistical distances representing different methodologies: Using low-dimensional projections (Sliced-Wasserstein; SW), obtaining a distance using classifiers (Classifier Two-Sample Tests; C2ST), using embeddings through kernels (Maximum Mean Discrepancy; MMD), or neural networks (Fréchet Inception Distance; FID). We highlight the intuition behind each distance and explain their merits, scalability, complexity, and pitfalls. To demonstrate how these distances are used in practice, we evaluate generative models from different scientific domains, namely a model of decision making and a model generating medical images. We showcase that distinct distances can give different results on similar data. Through this guide, we aim to help researchers to use, interpret, and evaluate statistical distances for generative models in science.
△ Less
Submitted 19 March, 2024;
originally announced March 2024.
-
All product eigenstates in Heisenberg models from a graphical construction
Authors:
Felix Gerken,
Ingo Runkel,
Christoph Schweigert,
Thore Posske
Abstract:
Recently, large degeneracy based on product eigenstates has been found in spin ladders, Kagome-like lattices, and motif magnetism, connected to spin liquids, anyonic phases, and quantum scars. We unify these systems by a complete classification of product eigenstates of Heisenberg XXZ Hamiltonians with Dzyaloshinskii-Moriya interaction on general graphs in the form of Kirchhoff rules for spin supe…
▽ More
Recently, large degeneracy based on product eigenstates has been found in spin ladders, Kagome-like lattices, and motif magnetism, connected to spin liquids, anyonic phases, and quantum scars. We unify these systems by a complete classification of product eigenstates of Heisenberg XXZ Hamiltonians with Dzyaloshinskii-Moriya interaction on general graphs in the form of Kirchhoff rules for spin supercurrent. By this, we construct spin systems with extensive degree of degeneracy linked to exotic condensates which can be studied in atomic gases and quantum spin lattices.
△ Less
Submitted 19 October, 2023;
originally announced October 2023.
-
Quantum spin helices more stable than the ground state: onset of helical protection
Authors:
Stefan Kühn,
Felix Gerken,
Lena Funcke,
Tobias Hartung,
Paolo Stornati,
Karl Jansen,
Thore Posske
Abstract:
Topological magnetic structures are promising candidates for resilient information storage. An elementary example are spin helices in one-dimensional easy-plane quantum magnets. To quantify their stability, we numerically implement the stochastic Schrödinger equation and time-dependent perturbation theory for spin chains with fluctuating local magnetic fields. We find two classes of quantum spin h…
▽ More
Topological magnetic structures are promising candidates for resilient information storage. An elementary example are spin helices in one-dimensional easy-plane quantum magnets. To quantify their stability, we numerically implement the stochastic Schrödinger equation and time-dependent perturbation theory for spin chains with fluctuating local magnetic fields. We find two classes of quantum spin helices that can reach and even exceed ground-state stability: Spin-current-maximizing helices and, for fine-tuned boundary conditions, the recently discovered "phantom helices". Beyond that, we show that the helicity itself (left- or right-rotating) is even more stable. We explain these findings by separated helical sectors and connect them to topological sectors in continuous spin systems. The resulting helical protection mechanism is a promising phenomenon towards stabilizing helical quantum structures, e.g., in ultracold atoms and solid state systems. We also identify an - up to our knowledge - previously unknown new type of phantom helices.
△ Less
Submitted 28 February, 2023; v1 submitted 6 February, 2023;
originally announced February 2023.
-
Unique Signatures of Topological Phases in Two-Dimensional THz Spectroscopy
Authors:
Felix Gerken,
Thore Posske,
Shaul Mukamel,
Michael Thorwart
Abstract:
We develop a microscopic theory for the two-dimensional spectroscopy of one-dimensional topological superconductors. We consider a ring geometry as a realization of the Kitaev chain with periodic boundary conditions. We show numerically and analytically that the cross-peak structure of the 2D spectra carries unique signatures of the topological phases of the chain. Our work reveals how 2D spectros…
▽ More
We develop a microscopic theory for the two-dimensional spectroscopy of one-dimensional topological superconductors. We consider a ring geometry as a realization of the Kitaev chain with periodic boundary conditions. We show numerically and analytically that the cross-peak structure of the 2D spectra carries unique signatures of the topological phases of the chain. Our work reveals how 2D spectroscopy can identify topological phases in bulk properties, bypassing energy-specific differences caused by topologically protected or trivial boundary modes that are otherwise hard to distinguish.
△ Less
Submitted 31 January, 2022;
originally announced February 2022.
