← 研踪 PaperTrailarXiv:2609.27964 · v3
Learning from Transients: Linear RNN Scaling Laws in Stationary Prediction
首次提交 2026-08-23 21:37
版本更新 2026-11-19 12:33
雷达观测 2026-11-20 10:50
摘要
Does the organization of a fixed number of transitions into independent trajectories affect a fixed stationary prediction task? We study a Gaussian-sketched linear recurrent teacher--student model, where independently initialized training trajectories create transients that can excite weak spectral directions. For full-batch gradient descent with stable deterministic learning rates, we establish matching stationary approximation, bias, variance, and total-risk scaling laws from dependent trajectory data. When initialization decays at least as fast spectrally as the innovations, trajectory count and length enter the leading variance rate through their product. With a heavier initialization tail, transient excitation can improve the trajectory-length-dependent finite-time bias and, at fixed transition count, variance, while the stationary oracle shift remains controlled by representation approximation. Under token and compute caps, constant-length trajectories attain the optimal risk scaling order within the theorem-controlled range. Synthetic experiments provide finite-scale evidence in this stylized model.
作者备注:65 pages, 4 figures
观测到的版本
相邻记录可能跳过中间版本;首次发现 v2 及以上也会标为版本更新。
v3 · 2026-11-19 12:33 · 最新观测版本
Learning from Transients: Linear RNN Scaling Laws in Stationary Prediction
Does the organization of a fixed number of transitions into independent trajectories affect a fixed stationary prediction task? We study a Gaussian-sketched linear recurrent teacher--student model, where independently initialized training trajectories create transients that can excite weak spectral directions. For full-batch gradient descent with stable deterministic learning rates, we establish matching stationary approximation, bias, variance, and total-risk scaling laws from dependent trajectory data. When initialization decays at least as fast spectrally as the innovations, trajectory count and length enter the leading variance rate through their product. With a heavier initialization tail, transient excitation can improve the trajectory-length-dependent finite-time bias and, at fixed transition count, variance, while the stationary oracle shift remains controlled by representation approximation. Under token and compute caps, constant-length trajectories attain the optimal risk scaling order within the theorem-controlled range. Synthetic experiments provide finite-scale evidence in this stylized model.
查看此版本解读 →查看此版本原文 ↗