Bringing the Discussion of Minima Sharpness to the Audio Domain: a Filter-Normalised Evaluation for Acoustic Scene Classification

TL;DR

This paper investigates the relationship between loss minima sharpness and generalisation in acoustic scene classification, revealing sharper minima can sometimes generalise better, especially for out-of-domain data, challenging previous assumptions.

Contribution

It introduces a filter-normalised visualisation and sharpness measure for acoustic scene classification, showing sharper minima can outperform flat ones in generalisation, and discusses optimizer effects.

Findings

Sharper minima often lead to better out-of-domain generalisation.

Optimizer choice significantly influences minima sharpness.

Publicly available code and visualisations support reproducibility.

Abstract

The correlation between the sharpness of loss minima and generalisation in the context of deep neural networks has been subject to discussion for a long time. Whilst mostly investigated in the context of selected benchmark data sets in the area of computer vision, we explore this aspect for the acoustic scene classification task of the DCASE2020 challenge data. Our analysis is based on two-dimensional filter-normalised visualisations and a derived sharpness measure. Our exploratory analysis shows that sharper minima tend to show better generalisation than flat minima -even more so for out-of-domain data, recorded from previously unseen devices-, thus adding to the dispute about better generalisation capabilities of flat minima. We further find that, in particular, the choice of optimisers is a main driver of the sharpness of minima and we discuss resulting limitations with respect to comparability. Our code, trained model states and loss landscape visualisations are publicly available.