PACT：视觉语言模型基于适配器的持续学习中的类相变约束
PACT: Phase-Like Transition Constraints in Adapter-Based Continual Learning of Vision-Language Models
在基于适配器的持续学习中，用类似相变的稳定性/可塑性正则与先验锚定替代正交约束，让视觉语言模型在任务间保留可迁移共享；做增量适配器训练时可复用其双分支 ViT 结构与稳定适配器初始化。
原文依据 we provide a new perspective based on PAC-Bayesian analysis: once the per-task optimization has converged, adapters should be further shaped
摘要译文
由模型翻译 · 术语统一 · 请对照原文核对关键结论持续学习（CL）使视觉语言模型（VLM）能够在保留已有知识的同时获得新能力，例如通过使用任务特定的适配器。现有的持续学习方法通常将这些适配器优化至收敛，并往往施加（近）正交约束以减少干扰；然而，将适配器隔离在正交子空间中会抑制跨任务迁移与共享。为解决这一问题，我们基于 PAC-贝叶斯分析提供了一种新视角：一旦逐任务优化已经收敛，就应进一步塑造适配器，使其满足类相变约束（PACT）——一种由两部分构成的表述，即（i）规定适配器之间的类相变关系，（ii）施加显式约束以强制实现该关系。在 PACT 下，适配器动态类似于水的相变：系统倾向于进入“冻结”（保留历史、约束紧密）或“融化”（任务自适应、自由）状态，并在两者之间平滑移动，而非通过硬阈值切换。我们通过在双分支视觉 Transformer（ViT）中耦合稳定性与可塑性正则项、以稳定适配器初始化（SAI）为适配器赋初值，并引入先验锚定（PA）机制来落实 PACT，从而诱导出类相变的适配器动态。在多种持续学习设置下，PACT 超越了最优方法，同时相对于标准适配器基线将可训练参数数量减少了 36.96%。我们的代码将公开发布。
英文原文摘要 1592 字符
Continual Learning (CL) enables Vision-Language Models (VLMs) to acquire new capabilities while retaining prior knowledge, for example, by employing task-specific adapters. Existing CL approaches typically optimize these adapters to convergence, often with (near-)orthogonality constraints to reduce interference; however, isolating adapters in orthogonal subspaces can suppress cross-task transfer and sharing. To address this problem, we provide a new perspective based on PAC-Bayesian analysis: once the per-task optimization has converged, adapters should be further shaped to satisfy \underline P hase-like tr\underline A nsition \underline C ons\underline T raints (PACT) -- a two-part formulation that (i) specifies a phase-like transition relation among adapters and (ii) imposes explicit constraints that enforce this relation. Under PACT, adapter dynamics resemble the phase transition of water: the system gravitates toward either a "frozen" (history-preserving, tightly constrained) or a "melted" (task-adaptive, free) regime, while moving between them smoothly rather than via hard thresholds. We operationalize PACT by coupling stability and plasticity regularizers within a two-branch Vision Transformer (ViT), seeding adapters with a Stable Adapter Initialization (SAI), and introducing a Prior Anchoring (PA) mechanism, thereby inducing phase-like adapter dynamics. Across diverse CL settings, PACT surpasses state-of-the-art methods while reducing the number of trainable parameters by 36.96% relative to standard adapter-based baselines. Our code will be released publicly.