其实是对着 UMich EECS 498-007 / 598-005: Deep Learning for Computer Vision 写的，所以主题是 Computer Vision 。本文中会有配图，是从 Justin Johnson 的 slides 上蒯的。
-
维护日志
编撰以应对并记录可能出现的技术难处。
-
Bellman Equations
Bellman Equations are basic tools to characterize MDPs and MRPs. Here we aim to lay theoritcal foundations for general MDPs. While sometimes it's necessary and convenience to impose certain constraints, e.g. polish space or discrete space, to ensure regularity.
-
Markov Decision Processes, Policy Evaluation, and RL Terminologies
Markov Decision Process is the threotical foundation of reincement learning. This installment introduces the foundamental settings of Reinforcement Learning.
-
Discussion to Markov Properties
We know that Markov properties are used in extremely wide range of areas. But it has always been lack of understanding for me. Here we aim to provide as rigorous an introduction as possible to the ... -
Discussion to Differetials and Derivatives
我以前一直没怎么弄明白微分，而且现在越来越多的接触到矩阵微分，经常碰到定义上和计算上的问题，所以我们要在这里重新介绍微分那一套。特别的，比较严格的理论（主要是微分的某些内蕴性质）恐怕要在微分几何上才会遇到，因此我们暂时不会涉及这一部分。本文的主线是一般流形上的一阶微分，并以矩阵空间为例。
-
Basic Concepts of Covariance
由于本人只学过概率论，所以在这里简单的复习一点协方差的意义和性质。
-
Variational Auto Encoder
本文将简要的介绍 VAE 。本来还想谈谈 VQ-VAE ，但是觉得这种模型比较平凡，因此不讲了。
-
Factor Analysis and Principal Components Analysis
本文原先是要讲 Mixture of Gaussians, Factor Analysis, Independent Components Analysis 和 Principal Components Analysis 的。但后来我觉得 MoG 在计算上实在是没什么好讲的，而 ICA 的假设太强而脱离实际，所以这两者就不讲了。