CS代考 https://xkcd.com/835/

https://xkcd.com/835/
Announcements:
● Guest lecture this Wed
PHYS T + Teams

Copyright By PowCoder代写 加微信 powcoder

ML for cyber security
See you here!
Two graphical model tutorials: this
week and next week.
Hope you’re going well in assignment 2
Final exam Fri 3 June 5:40 – 8:40 pm Canberra
○ On Wattle
○ Self-invigilated, video
recording needed ○ Instructions will be

Graphical model inference
Factor graphs
The sum-product algorithm
Other algorithms
High-level goal:
One representation for both directed and
this representation to facilitate inference.
[Frey, 1998, Kschischnang et al 2001]
undirected graphical models.
Bishop 8.4.3, 8.4.4

Factor graphs
s: a subset of variables
Joint distribution → a product of factors
Contains more info than MRF, because fa(x1, x2)
and fb(x1, x2) would be in one potential function.
Directed graph: conditionals
Undirected graph: potential functions over max cliques→ factors
Z is a factor too, over an empty set of variables.

graphs representing
same distribution

graphs representing
same distribution

Factor graphs are

Factor graph for a poly-tree
Factor graphs for directed/undirected trees, and directed poly trees retain a tree structure.

factor graphs representing the same (undirected) graph
Factorization in (c) does not correspond to any conditional independence properties.

The sum-product algorithm
Goal: evaluate local marginals over nodes or subsets of nodes. Variant: find the most probable state → max sum algorithm.
Assume (for instructional convenience): all variables
For continuous variables – perform integration, e.g.
Historic note:
belief propagation (Pearl, 1988; Lauritzen and Spiegelhalter,
inference on DAGs. It is a special case of the sum product algorithm (Frey, 1998;
Kschischnang et al., 2001)
discrete. Marginalization = sums
linear dynamic systems (e.g. Chap 13.3)

Recall from last lecture …

The sum-product algorithm
s: a subset of nodes
ne(x) : neighbouring factor nodes
Xs: the set of all variables in the subtree connected to
the variable node x via the factor node fs Fs(x,Xs): the product of all the factors in the group associated with factor fs.

A first example of
sum-product

Factorise this
F (x,X ): the product of all the factors in the group associated with factor f .
factor subgraph:
ne(fs) : neighbours of factor node fs
ne(fs)\x : neighbours of factor node fs except x
Xs: the set of all variables in the subtree connected to the variable node x via the factor node fs
= {x, x1, … , x

Two different
kinds of messages:
Computing factor-to-variable-message
● take the product of the incoming messages along all other links coming into the factor
● multiply by the factor associated with that node
● marginalize over all of the variables associated with the incoming messages
can send a message once the factor node has received incoming
messages from all
other neighbouring variable nodes
{x1, … , xM

start, how
to normalise?

Aji, S.M.; McEliece, R.J. (Mar 2000). “The generalized distributive law”. IEEE Transactions Information Theory. 46 (2): 325–343. doi:10.1109/18.825794.

Graphical model inference
Factor graphs
The sum-product algorithm
Other algorithms

程序代写 CS代考 加微信: powcoder QQ: 1823890830 Email: powcoder@163.com