通过梯度手术实现低秩适配器初始化用于持续学习
Low-Rank Adapters Initialization via Gradient Surgery for Continual Learning
AI总结：
本文提出SLICE方法，通过梯度手术优化LoRA适配器初始化，提升持续学习中稳定性与可塑性的平衡，改进性能指标。
摘要：
LoRA因其参数效率、任务模块化和与重放策略的兼容性而广泛用于大型语言模型的持续微调。然而，基于LoRA的持续学习仍易受灾难性遗忘影响，其严重程度取决于连续任务梯度的交互：当连续任务梯度冲突时，标准适配器初始化会将更新引导至子空间，覆盖先前学习的方向。本文提出SLICE，一种基于梯度手术的LoRA适配器初始化方法。SLICE累计当前任务和先前任务重放缓冲区的梯度，通过投影算子进行协调，并通过截断SVD分解结果以初始化适配器权重。我们在TRACE基准和Super-NI任务序列上评估SLICE，包括我们通过挖掘梯度对立最显著的任务对构造的一组对抗性Super-NI序列。与vanilla LoRA、LoRA-GA和LoRAM相比，SLICE在持续学习序列中始终实现了更好的稳定性-可塑性平衡，提高平均性能、最终性能和遗忘指标，同时在标准和对抗性持续学习序列中保持通用性能和上下文性能。
英文摘要：
LoRA is widely adopted for continual fine-tuning of Large Language Models due to its parameter efficiency, modularity across tasks, and compatibility with replay strategies. However, LoRA-based continual learning remains vulnerable to catastrophic forgetting, whose severity depends on how successive task gradients interact: when consecutive task gradients conflict, standard adapter initializations channel updates into subspaces that overwrite previously learned directions. We propose SLICE, a gradient-surgery-based initialization for LoRA adapters in continual learning. SLICE accumulates gradients from both the current task and a replay buffer of prior tasks, reconciles them through a projection operator, and decomposes the result via truncated SVD to initialize the adapter weights. We evaluate SLICE on the TRACE benchmark and sequences of Super-NI tasks, including a set of adversarial Super-NI sequences that we construct by mining task pairs with maximally opposing gradients. Compared to vanilla LoRA, LoRA-GA, and LoRAM, SLICE consistently achieves a better stability-plasticity trade-off, improving Average Performance, Final Performance and Forgetting metrics while preserving General Performance and In Context Performance across both standard and adversarial continual learning sequences.
机构、总结、中文摘要均由AI分析而来，可能存在错误，请以论文正文为准。
