用神经正切核动态选点，训练隐式神经表示快一倍
NTK-Guided Implicit Neural Teaching
- 基于神经正切核筛选最具影响力的坐标点
- 训练时间减半，图像重建质量不降反升
- 适合需要高效训练隐式模型的研究者
隐式神经表示（INRs）通过多层感知机参数化连续信号，实现图像、音频和3D重建等任务的紧凑、分辨率无关建模。但高分辨率信号需优化数百万个坐标，计算成本极高。为此，我们提出NTK-Guided Implicit Neural Teaching（NINT），通过动态选择能最大化全局函数更新的坐标来加速训练。利用神经正切核（NTK），NINT根据损失梯度的NTK增强范数对样本评分，同时捕捉拟合误差与异质影响力（自影响与跨坐标耦合）。实验表明，NINT在保持或提升表示质量的同时，将训练时间减少近一半，成为当前基于采样的策略中加速效果最佳的方法。
原文摘要 · Abstract (English)
Implicit Neural Representations (INRs) parameterize continuous signals via multilayer perceptrons (MLPs), enabling compact, resolution-independent modeling for tasks like image, audio, and 3D reconstruction. However, fitting high-resolution signals demands optimizing over millions of coordinates, incurring prohibitive computational costs. To address it, we propose NTK-Guided Implicit Neural Teaching (NINT), which accelerates training by dynamically selecting coordinates that maximize global functional updates. Leveraging the Neural Tangent Kernel (NTK), NINT scores examples by the norm of their NTK-augmented loss gradients, capturing both fitting errors and heterogeneous leverage (self-influence and cross-coordinate coupling). This dual consideration enables faster convergence compared to existing methods. Through extensive experiments, we demonstrate that NINT significantly reduces training time by nearly half while maintaining or improving representation quality, establishing state-of-the-art acceleration among recent sampling-based strategies.
Thank you to arXiv for use of its open access interoperability. PaperDance 不是 arXiv 官方产品；中文卡片由大模型生成，请以原文为准。