Soft-TransFormers for Continual Learning

Fuente: arXiv
Salvato in:
Dettagli Bibliografici
Autori principali: Kang, Haeyong, Yoo, Chang D.
Natura: Preprint
Pubblicazione: 2024
Soggetti:
Accesso online:
Tags: Aggiungi Tag
Nessun Tag, puoi essere il primo ad aggiungerne!!
_version_ 1866914511555919872
author Kang, Haeyong
Yoo, Chang D.
author_facet Kang, Haeyong
Yoo, Chang D.
contents Inspired by the \emph{Well-initialized Lottery Ticket Hypothesis (WLTH)}, we introduce Soft-Transformer (Soft-TF), a parameter-efficient framework for continual learning that leverages soft, real-valued subnetworks over a frozen pre-trained Transformer. Instead of relying on manually designed prompts or adapters, Soft-TF learns task-specific multiplicative masks applied to the key, query, value, and output projections in self-attention. These masks enable smooth and stable task adaptation while preserving shared representations. Combined with a lightweight dual-prompt mechanism, Soft-TF maintains strong knowledge retention and mitigates Catastrophic Forgetting (CF). Across multiple continual learning benchmarks, Soft-TF achieves state-of-the-art performance, consistently outperforming prompt-based, adapter-based, and LoRA-style baselines while requiring minimal additional parameters.
format Preprint
id arxiv_https___arxiv_org_abs_2411_16073
institution arXiv
publishDate 2024
record_format arxiv
spellingShingle Soft-TransFormers for Continual Learning
Kang, Haeyong
Yoo, Chang D.
Machine Learning
Artificial Intelligence
Computer Vision and Pattern Recognition
Inspired by the \emph{Well-initialized Lottery Ticket Hypothesis (WLTH)}, we introduce Soft-Transformer (Soft-TF), a parameter-efficient framework for continual learning that leverages soft, real-valued subnetworks over a frozen pre-trained Transformer. Instead of relying on manually designed prompts or adapters, Soft-TF learns task-specific multiplicative masks applied to the key, query, value, and output projections in self-attention. These masks enable smooth and stable task adaptation while preserving shared representations. Combined with a lightweight dual-prompt mechanism, Soft-TF maintains strong knowledge retention and mitigates Catastrophic Forgetting (CF). Across multiple continual learning benchmarks, Soft-TF achieves state-of-the-art performance, consistently outperforming prompt-based, adapter-based, and LoRA-style baselines while requiring minimal additional parameters.
title Soft-TransFormers for Continual Learning
topic Machine Learning
Artificial Intelligence
Computer Vision and Pattern Recognition
url https://arxiv.org/abs/2411.16073