Current browse context:
cs.LG
Change to browse by:
References & Citations
Computer Science > Machine Learning
Title: Instabilities in Convnets for Raw Audio
(Submitted on 11 Sep 2023 (v1), last revised 26 Apr 2024 (this version, v4))
Abstract: What makes waveform-based deep learning so hard? Despite numerous attempts at training convolutional neural networks (convnets) for filterbank design, they often fail to outperform hand-crafted baselines. These baselines are linear time-invariant systems: as such, they can be approximated by convnets with wide receptive fields. Yet, in practice, gradient-based optimization leads to suboptimal approximations. In our article, we approach this phenomenon from the perspective of initialization. We present a theory of large deviations for the energy response of FIR filterbanks with random Gaussian weights. We find that deviations worsen for large filters and locally periodic input signals, which are both typical for audio signal processing applications. Numerical simulations align with our theory and suggest that the condition number of a convolutional layer follows a logarithmic scaling law between the number and length of the filters, which is reminiscent of discrete wavelet bases.
Submission history
From: Daniel Haider [view email][v1] Mon, 11 Sep 2023 22:34:06 GMT (215kb,D)
[v2] Sat, 21 Oct 2023 10:54:08 GMT (1232kb,D)
[v3] Tue, 16 Apr 2024 11:40:46 GMT (1105kb,D)
[v4] Fri, 26 Apr 2024 08:25:12 GMT (1105kb,D)
Link back to: arXiv, form interface, contact.