-
「A proof of concept tool to verify estimates」笔记：问题从哪来、证明卡在哪
This post was inspired by some recent discussions with Bjoern Bringmann.…
-
2026Transformer自回归推理优化指南：从O(n³)到O(n²)降本落地技巧
未经优化的Transformer自回归推理存在O(n³)超高复杂度，是大模型卡顿、token长度受限的核心根源。通过各注意力层增量缓存优化，可将…
-
「Orders of infinity」笔记：问题从哪来、证明卡在哪
Many problems in analysis (as well as adjacent fields such as combinator…
-
「A tool to verify estimates, II: a flexible proof」：数学直觉如何落到可检验的步骤
In a recent post, I talked about a proof of concept tool to verify estim…
-
深度学习自动微分正反模式详解：2026落地选型与算力优化技巧
深度学习框架的梯度计算核心依托自动微分（AD）技术，主流框架默认采用反向模式自动微分，而非大众误解的前向传播。前向、反向两种自动微分模式基于链式…
-
「Some variants of the periodic tiling conjecture」笔记：问题从哪来、证明卡在哪
Rachel Greenfeld and I have just uploaded to the arXiv our paper Some va…
-
读懂「A Lean companion to “Analysis I”」：先把定义、反例和适用范围钉死
Almost 20 years ago, I wrote a textbook in real analysis called “Analysi…
-
2026实测Transformer自回归推理优化方案：从O(n³)降维O(n²)落地教程
未优化的Transformer自回归推理存在O(n³)超高复杂度，是大模型解码卡顿、长文本受限的核心原因。通过注意力层增量计算+Key/Valu…
-
读懂「On the number of exceptional intervals to the pr」：先把定义、反例和适用范围钉死
Ayla Gafni and I have just uploaded to the arXiv the paper “On the numbe…
-
把「Decomposing a factorial into large factors (seco」写成可执行的阅读路径
Boris Alexeev, Evan Conway, Matthieu Rosenfeld, Andrew Sutherland, Marku…