Pruning as Regularization: Sensitivity-Aware One-Shot Pruning in ASR

Fuente: arXiv
Guardado en:
Detalles Bibliográficos
Autores principales: Irigoyen, Julian, Söhler, Arthur, Kirkedal, Andreas Søeborg
Formato: Preprint
Publicado: 2025
Materias:
Acceso en línea:
Etiquetas: Agregar Etiqueta
Sin Etiquetas, Sea el primero en etiquetar este registro!
_version_ 1866912701725278208
author Irigoyen, Julian
Söhler, Arthur
Kirkedal, Andreas Søeborg
author_facet Irigoyen, Julian
Söhler, Arthur
Kirkedal, Andreas Søeborg
contents We challenge the conventional view of neural network pruning as solely a compression technique, demonstrating that one-shot magnitude pruning serves as a powerful implicit regularizer for ASR. Using Whisper-small, we combine gradient- and Fisher-based sensitivity diagnostics with targeted, component-wise pruning. This reveals architectural asymmetries: decoder FFNs are pruning-fragile, whereas decoder self-attention and the last encoder layers contain redundancy that, when removed, improves generalization. Without fine-tuning, pruning 50% of decoder self-attention reduces WER by 2.38% absolute (20.44% relative) on LibriSpeech test-other; pruning the last four encoder layers at 50% instead yields a 1.72% absolute (14.8% relative) improvement. Gains persisted on Common Voice and TED-LIUM datasets. Beyond regularization benefits, our sensitivity-aware approach enables more aggressive one-shot compression. At 40% sparsity, where established global pruning approaches catastrophically fail, our method preserves near-baseline accuracy. This positions pruning as a first-class architectural design tool: knowing where to prune is as important as how much to prune.
format Preprint
id arxiv_https___arxiv_org_abs_2511_08092
institution arXiv
publishDate 2025
record_format arxiv
spellingShingle Pruning as Regularization: Sensitivity-Aware One-Shot Pruning in ASR
Irigoyen, Julian
Söhler, Arthur
Kirkedal, Andreas Søeborg
Audio and Speech Processing
Computation and Language
Sound
We challenge the conventional view of neural network pruning as solely a compression technique, demonstrating that one-shot magnitude pruning serves as a powerful implicit regularizer for ASR. Using Whisper-small, we combine gradient- and Fisher-based sensitivity diagnostics with targeted, component-wise pruning. This reveals architectural asymmetries: decoder FFNs are pruning-fragile, whereas decoder self-attention and the last encoder layers contain redundancy that, when removed, improves generalization. Without fine-tuning, pruning 50% of decoder self-attention reduces WER by 2.38% absolute (20.44% relative) on LibriSpeech test-other; pruning the last four encoder layers at 50% instead yields a 1.72% absolute (14.8% relative) improvement. Gains persisted on Common Voice and TED-LIUM datasets. Beyond regularization benefits, our sensitivity-aware approach enables more aggressive one-shot compression. At 40% sparsity, where established global pruning approaches catastrophically fail, our method preserves near-baseline accuracy. This positions pruning as a first-class architectural design tool: knowing where to prune is as important as how much to prune.
title Pruning as Regularization: Sensitivity-Aware One-Shot Pruning in ASR
topic Audio and Speech Processing
Computation and Language
Sound
url https://arxiv.org/abs/2511.08092