Variational Inference

Variational Inference trace a upper bound on the, mostly untraceable, Free Energy.

---

$=D_{KL}[q(H|D;\theta )\Vert p(H|D)]-\ln p(D)$ $\ge D_{KL}[q(H|D;\theta )\Vert p(H|D)]$ commonly written as: $F[Q,y]=\underset{Divergence}{\underbrace{D_{KL}[Q(x)\Vert P(x|y)]}}-\underset{Evidence}{\underbrace{\ln P(y)}}$

---

Evidence

Supporting

Handle	Title	Summary of relevant evidence	Link
Millidge et al. 2021	A Mathematical Walkthrough and Discussion of the Free Energy Principle	p. 11
Parr, Pezzulo, Friston 2022	Active Inference Textbook configuration.	p. 28 & 30

Counter

Handle	Title	Summary of relevant evidence	Link

---

Connections

Used to approximate Self-evidencing Used to trace and minimize the Variational Free Energy The Free-Energy Principle

Can be recomposed to and

---

Metadata

---

Sentence

• WiP