Scribd
Upload
24 views17 pages

MCMC - Silver Bullet of Probabilistic Modeling

MCMC is useful for probabilistic modeling by estimating expected values through sampling. The document discusses how MCMC can be applied to Bayesian inference and the EM algorithm by estimating integrals over distributions that are difficult to directly compute.

Uploaded by

Valeria
Copyright
© © All Rights Reserved
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as PDF, TXT or read online on Scribd
24 views17 pages

MCMC - Silver Bullet of Probabilistic Modeling

MCMC is useful for probabilistic modeling by estimating expected values through sampling. The document discusses how MCMC can be applied to Bayesian inference and the EM algorithm by estimating integrals over distributions that are difficult to directly compute.

Uploaded by

Valeria
Copyright
© © All Rights Reserved
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as PDF, TXT or read online on Scribd
You are on page 1/ 17

Markov Chain Monte Carlo (MCMC)

• MCMC — silver bullet of probabilistic modeling


Markov Chain Monte Carlo (MCMC)

• MCMC — silver bullet of probabilistic modeling

• Learn how exploit specifics of your problem to


speed up MCMC
Markov Chain Monte Carlo (MCMC)

• MCMC — silver bullet of probabilistic modeling

• Learn how exploit specifics of your problem to


speed up MCMC

• Understand the limitations


Monte Carlo
Monte Carlo
Monte Carlo

Estimate expected values by sampling


Monte Carlo

Estimate expected values by sampling

⇡
= E [x2 + y 2  1]
4
Monte Carlo

Estimate expected values by sampling

⇡
= E [x2 + y 2  1]
4
M
X
1
⇡ [x2s + ys2  1]
M s=1

xs , ys ⇠ U(0, 1)
Monte Carlo

Estimate expected values by sampling

XM
1
Ep(x) f (x) ⇡ f (xs )
M s=1
Monte Carlo

Estimate expected values by sampling

XM
1
Ep(x) f (x) ⇡ f (xs )
M s=1

xs ⇠ p(x)
Monte Carlo

Why do we need to estimate expected values?


Monte Carlo

Why do we need to estimate expected values?

• Full Bayesian inference (see Week 1)

p(y | x, Ytrain , Xtrain )


Z
= p(y | x, w)p(w | Ytrain , Xtrain )dw

= Ep(w|Ytrain ,Xtrain ) p(y | x, w)


Monte Carlo

Why do we need to estimate expected values?

• Full Bayesian inference (see Week 1)

p(y | x, Ytrain , Xtrain )


Z
= p(y | x, w)p(w | Ytrain , Xtrain )dw

= Ep(w|Ytrain ,Xtrain ) p(y | x, w)


Monte Carlo

Why do we need to estimate expected values?

• Full Bayesian inference (see Week 1)

p(y | x, Ytrain , Xtrain )


Z
= p(y | x, w)p(w | Ytrain , Xtrain )dw

= Ep(w|Ytrain ,Xtrain ) p(y | x, w)


Monte Carlo

Why do we need to estimate expected values?

• Full Bayesian inference (see Week 1)

p(y | x, Ytrain , Xtrain )


Z
= p(y | x, w)p(w | Ytrain , Xtrain )dw

= Ep(w|Ytrain ,Xtrain ) p(y | x, w)

p(Ytrain | Xtrain , w)p(w)


p(w | Ytrain , Xtrain ) =
Z
Monte Carlo

Why do we need to estimate expected values?

• Full Bayesian inference (see Week 1)

• M-step of EM-algorithm (see Week 2)


Monte Carlo

Why do we need to estimate expected values?

• Full Bayesian inference (see Week 1)

• M-step of EM-algorithm (see Week 2)

max. Eq log p(X, T | ✓)


✓

You might also like