https://kino204.github.io/kINo's BlogsNotes & ideas. 2026-09-27T17:33:55+08:00 kino https://kino204.github.io/ Jekyll © 2026 kino /assets/img/favicons/favicon.ico /assets/img/favicons/favicon-96x96.png Automatic Differentiation2026-09-27T00:00:00+08:00 2026-09-27T17:33:29+08:00 https://kino204.github.io/posts/autodiff/ kino Automatic differentiation(AD) is one of the core techniques in modern deep learning frameworks. There are two main methods: forward & reverse mode AD. The Chain Rule Chain rule is the core principle for both forward & reverse AD. The working environment for AD is a calculation DAG, with each node being a tensor $\mathbb{R}^{n}$. Take two node $x$ & $y$, given that there exists a p... Conditional Probability as a Fully-connected Probability Model2026-09-07T10:51:00+08:00 2026-09-14T14:27:56+08:00 https://kino204.github.io/posts/conditional-probability-as-a-fully-connected-probability-model/ kino Markov condition [p(x_{i} \Omega \backslash \mathbf{de}(x_{i})) = p(x_{i} \mathbf{pa}(x_{i}))] provides prior belief about “relevancy” between events.