Efficiently Train ASR Models that Memorize Less and Perform Better with Per-core Clipping

Fuente: arXiv
Enregistré dans:
Détails bibliographiques
Auteurs principaux: Wang, Lun, Thakkar, Om, Meng, Zhong, Rafidi, Nicole, Prabhavalkar, Rohit, Narayanan, Arun
Format: Preprint
Publié: 2024
Sujets:
Accès en ligne:
Tags: Ajouter un tag
Pas de tags, Soyez le premier à ajouter un tag!
_version_ 1866911907647062016
author Wang, Lun
Thakkar, Om
Meng, Zhong
Rafidi, Nicole
Prabhavalkar, Rohit
Narayanan, Arun
author_facet Wang, Lun
Thakkar, Om
Meng, Zhong
Rafidi, Nicole
Prabhavalkar, Rohit
Narayanan, Arun
contents Gradient clipping plays a vital role in training large-scale automatic speech recognition (ASR) models. It is typically applied to minibatch gradients to prevent gradient explosion, and to the individual sample gradients to mitigate unintended memorization. This work systematically investigates the impact of a specific granularity of gradient clipping, namely per-core clip-ping (PCC), across training a wide range of ASR models. We empirically demonstrate that PCC can effectively mitigate unintended memorization in ASR models. Surprisingly, we find that PCC positively influences ASR performance metrics, leading to improved convergence rates and reduced word error rates. To avoid tuning the additional hyperparameter introduced by PCC, we further propose a novel variant, adaptive per-core clipping (APCC), for streamlined optimization. Our findings highlight the multifaceted benefits of PCC as a strategy for robust, privacy-forward ASR model training.
format Preprint
id arxiv_https___arxiv_org_abs_2406_02004
institution arXiv
publishDate 2024
record_format arxiv
spellingShingle Efficiently Train ASR Models that Memorize Less and Perform Better with Per-core Clipping
Wang, Lun
Thakkar, Om
Meng, Zhong
Rafidi, Nicole
Prabhavalkar, Rohit
Narayanan, Arun
Cryptography and Security
Computation and Language
Sound
Audio and Speech Processing
Gradient clipping plays a vital role in training large-scale automatic speech recognition (ASR) models. It is typically applied to minibatch gradients to prevent gradient explosion, and to the individual sample gradients to mitigate unintended memorization. This work systematically investigates the impact of a specific granularity of gradient clipping, namely per-core clip-ping (PCC), across training a wide range of ASR models. We empirically demonstrate that PCC can effectively mitigate unintended memorization in ASR models. Surprisingly, we find that PCC positively influences ASR performance metrics, leading to improved convergence rates and reduced word error rates. To avoid tuning the additional hyperparameter introduced by PCC, we further propose a novel variant, adaptive per-core clipping (APCC), for streamlined optimization. Our findings highlight the multifaceted benefits of PCC as a strategy for robust, privacy-forward ASR model training.
title Efficiently Train ASR Models that Memorize Less and Perform Better with Per-core Clipping
topic Cryptography and Security
Computation and Language
Sound
Audio and Speech Processing
url https://arxiv.org/abs/2406.02004