面向大脑编码的参数高效微调扩展研究
原标题：Paper page - Scaling Parameter-Efficient Fine-Tuning for Brain Encoding
AI 摘要
该论文提出用参数高效微调（PEFT）与低秩适配器（LoRA）将超大多模态骨干模型适配到大脑编码任务，通过利用骨干的因果结构单次预测多个时间点，显著降低特征提取成本，从而可高效微调最多 104B 参数的模型。方法 BrainLoRA 在 Algonauts 2025 挑战赛上取得已发表最佳成绩，并借助自研 multi-LoRA 内核训练 200+ 适配模型集成，在公开赛后榜单排名第一。研究还表明适配后的多模态骨干保持可解释性，神经适配提升了音频输入与跨模态混合的贡献。
正文节选
Scaling Parameter-Efficient Fine-Tuning for Brain Encoding Abstract Parameter-efficient fine-tuning with low-rank adapters enables efficient adaptation of very large multimodal backbones for brain encoding, improving performance and interpretability while reducing computational cost. Brain encoding models aim to predict neural responses to stimuli. The most widely used approach fits readouts from frozen pretrained backbones. Using neural data to train large models from scratch or fully fine-tune