Search:
Match:
2 results

Analysis

This paper addresses a fundamental issue in the analysis of optimization methods using continuous-time models (ODEs). The core problem is that the convergence rates of these ODE models can be misleading due to time rescaling. The paper introduces the concept of 'essential convergence rate' to provide a more robust and meaningful measure of convergence. The significance lies in establishing a lower bound on the convergence rate achievable by discretizing the ODE, thus providing a more reliable way to compare and evaluate different optimization methods based on their continuous-time representations.
Reference

The paper introduces the notion of the essential convergence rate and justifies it by proving that, under appropriate assumptions on discretization, no method obtained by discretizing an ODE can achieve a faster rate than its essential convergence rate.

Analysis

This paper introduces novel generalizations of entanglement entropy using Unit-Invariant Singular Value Decomposition (UISVD). These new measures are designed to be invariant under scale transformations, making them suitable for scenarios where standard entanglement entropy might be problematic, such as in non-Hermitian systems or when input and output spaces have different dimensions. The authors demonstrate the utility of UISVD-based entropies in various physical contexts, including Biorthogonal Quantum Mechanics, random matrices, and Chern-Simons theory, highlighting their stability and physical relevance.
Reference

The UISVD yields stable, physically meaningful entropic spectra that are invariant under rescalings and normalisations.