Notes

Here I share my study notes on mathematics, computer vision, and related topics.

生成模型(1): Flow Matching
May 19, 2026

从 ODE 与向量场的视角构建生成模型:通过构造连接先验与数据分布的概率路径(高斯/ConOT 路径),利用条件流匹配损失等价训练神经网络速度场,实现从标准高斯噪声到真实数据样本的确定性转换,并推导了预测速度、噪声与原始数据三种等价的参数化形式。

Rethink Matrix
Jan 4, 2026

Beyond mere arrays of numbers, matrices represent linear transformations between vector spaces. This note formalizes my intuition on this topic.

← Back to home