SlideShare ist ein Scribd-Unternehmen logo
1 von 29
Downloaden Sie, um offline zu lesen
Dr. Florian Wilhelm
March 13th 2016
PyData Amsterdam
P 1
2
Whatโ€˜s the best model to
describe our data?
And what does โ€žbestโ€œ actually mean?
3
4
5
Simple model
โ€žGeneralityโ€œ
Complex model
โ€žBest Fitโ€œ
Occamโ€˜s Razor:
โ€žIt is vain to do with more
what can be done with fewerโ€œ
6
Simple Model
โ„‹1
Complex model
โ„‹2
Space of all possible datasets ๐ท
7
Simple Model
โ„‹1
Complex model
โ„‹2
โ„‹1 fits only a small subset of ๐ท well
8
Simple Model
โ„‹1
Complex model
โ„‹2
โ„‹2 can fit large parts of ๐ท well
9
Prefer the model with high evidence for a given dataset
Source: D. J. C. MacKay. Bayesian Interpolation. 1992
1. Model fitting: Assume โ„‹๐‘– is the right model and fit its parameters ๐’˜ with Bayes:
๐‘ƒ ๐’˜ ๐ท, โ„‹๐‘– =
๐‘ƒ ๐ท ๐’˜, โ„‹๐‘– ๐‘ƒ(๐’˜|โ„‹๐‘–)
๐‘ƒ(๐ท|โ„‹๐‘–)
โ€œBusiness as usualโ€
2. Model comparison: Compare different models with the help of their evidence
๐‘ƒ ๐ท โ„‹๐‘– and model prior ๐‘ƒ โ„‹๐‘– :
๐‘ƒ โ„‹๐‘– ๐ท โˆ ๐‘ƒ ๐ท โ„‹๐‘– ๐‘ƒ โ„‹๐‘–
โ€œOccamโ€˜s razor at workโ€œ
10
Marginalize & approximate:
๐‘ƒ ๐ท โ„‹๐‘– = ๐‘ƒ ๐ท ๐’˜, โ„‹๐‘– ๐‘ƒ ๐’˜ โ„‹๐‘– ๐‘‘๐’˜
๐‘ƒ ๐ท โ„‹๐‘– โ‰… ๐‘ƒ ๐ท ๐’˜ ๐‘€๐‘ƒ, โ„‹๐‘– ๐‘ƒ ๐’˜ ๐‘€๐‘ƒ โ„‹๐‘– โˆ†๐’˜
๐‘’๐‘ฃ๐‘–๐‘‘๐‘’๐‘›๐‘๐‘’ โ‰… ๐‘๐‘’๐‘ ๐‘ก ๐‘“๐‘–๐‘ก ๐‘™๐‘–๐‘˜๐‘’๐‘™๐‘–โ„Ž๐‘œ๐‘œ๐‘‘ ร— ๐‘‚๐‘๐‘๐‘Ž๐‘š ๐‘“๐‘Ž๐‘๐‘ก๐‘œ๐‘Ÿ
11
Occam factor:
โˆ†๐’˜
โˆ† ๐ŸŽ ๐’˜
Source: D. J. C. MacKay. Bayesian Interpolation. 1992
12
Given:
๏‚ง Dataset ๐ท = ๐‘ฅ ๐‘›, ๐‘ก ๐‘› with ๐‘› = 1 โ€ฆ ๐‘
๏‚ง Set of (non-linear) functions ฮฆ = {๐œ™โ„Ž: ๐‘ฅ โŸผ ๐œ™(๐‘ฅ)} with โ„Ž = 1 โ€ฆ ๐‘€
Assumption:
๐‘ฆ ๐’™; ๐’˜ =
โ„Ž=1
๐‘€
๐‘คโ„Ž ๐œ™โ„Ž(๐’™) ,
๐‘ก ๐‘› = ๐‘ฆ ๐’™; ๐’˜ + ๐œ ๐‘›,
where ๐œ ๐‘› is an additive noise with ๐’ฉ 0, ๐›ผโˆ’1
Task: Find min
๐’˜
โ€–ฮฆ๐’˜ โˆ’ ๐’•โ€–2
(Ordinary Least Squares)
13
14
Problem:
Having too many features leads to overfitting!
Regularization
Assumption: โ€žWeights are smallโ€œ
๐‘ ๐’˜; ๐œ† ~๐’ฉ(0, ๐œ†โˆ’1 ๐•€)
Task: Given ๐›ผ, ๐œ† find
min
๐’˜
๐›ผ ฮฆ๐’˜ โˆ’ ๐’• 2 + ๐œ† ๐’˜ 2
15
Consider each ๐›ผ๐‘–, ๐œ†๐‘– defining a model โ„‹๐‘– ๐›ผ, ๐œ† .
Yes! That means we can use
our Bayesian Interpolation to
find ๐’˜, ๐œถ, ๐€ with the highest
evidence!
This is the idea behind BayesianRidge as found in sklearn.linear_model
Consider that each weight has an individual variance, so that
๐‘ ๐’˜ ๐€ ~๐’ฉ 0, ฮ›โˆ’1 ,
where ฮ› = diag(๐œ†1, โ€ฆ , ๐œ† ๐ป), ๐œ†โ„Ž โˆˆ โ„+.
Now, our minimization problem is:
min
๐’˜
๐›ผ ฮฆ๐’˜ โˆ’ ๐’• 2 + ๐’˜ ๐‘กฮ›๐’˜
16
Pruning: If precision ๐œ†โ„Ž of feature โ„Ž is high, its weight ๐‘คโ„Ž is very likely to
be close to zero and is therefore pruned.
This is called Sparse Bayesian Learning or Automatic Relevance
Determination. Found as ARDRegression under sklearn.linear_model.
Crossvalidation can be used for the estimation of hyperparmeters but suffers from
the curse of dimensionality (inappropriate for low-statistics).
17
Source: Peter Ellerton, http://pactiss.org/2011/11/02/bayesian-inference-homo-bayesianis/
โ€ข Random 100 ร— 100 design matrix ฮฆ with 100 samples and 100
features
โ€ข Weights ๐‘ค๐‘–, ๐‘– โˆˆ ๐ผ = 1, โ€ฆ , 100 , random subset J โŠ‚ ๐ผ with ๐ฝ = 10, and
๐‘ค๐‘– =
0, ๐‘– โˆˆ ๐ผJ
๐’ฉ(๐‘ค๐‘–; 0, 1
4), ๐‘– โˆˆ ๐ฝ
โ€ข Target ๐’• = ฮฆ๐’˜ + ๐‚ with random noise ๐œˆ๐‘– โˆผ ๐’ฉ(0, 1
50)
Task: Reconstruct the weights, especially the 10 non-zero weights!
Source: http://scikit-learn.org/stable/auto_examples/linear_model/plot_ard.html#example-linear-model-plot-ard-py
18
19
20
21
22
We have to determine the parameters ๐‘ค, ๐œ†, ๐›ผ for
๐‘ƒ ๐’˜, ๐€, ๐›ผ ๐’• = ๐‘ƒ ๐’˜ ๐’•, ๐€, ๐›ผ ๐‘ƒ ๐€, ๐›ผ ๐’•
1) Model fitting:
For the first factor, we have ๐‘ƒ ๐’˜ ๐’•, ๐€, ๐›ผ ~๐’ฉ(๐, ฮฃ) with
ฮฃ = ฮ› + ๐›ผฮฆ ๐‘‡
ฮฆ โˆ’1
,
๐ = ๐›ผฮฃฮฆT ๐ญ.
23
2) Model comparison:
For the second factor, we have
๐‘ƒ ๐€, ๐›ผ ๐’• โˆ ๐‘ƒ ๐’• ๐€, ๐›ผ ๐‘ƒ ๐€ ๐‘ƒ ๐›ผ ,
where ๐‘ƒ ๐€ and ๐‘ƒ(๐›ผ) are hyperpriors which we assume uniform.
Using marginalization, we have
๐‘ƒ ๐’• ๐€, ๐›ผ = ๐‘ƒ ๐’• ๐’˜, ๐›ผ ๐‘ƒ ๐’˜ ๐€ ๐‘‘๐’˜,
i.e. marginal likelihood or the โ€œevidence for the hyperparameterโ€œ.
24
Differentiation of the log marginal likelihood with respect to ๐œ†๐‘– and ๐›ผ as
well as setting these to zero, we get
๐œ†๐‘– =
๐›พ๐‘–
๐œ‡๐‘–
2 ,
๐›ผ =
๐‘ โˆ’ ๐‘– ๐›พ๐‘–
๐’• โˆ’ ฮฆ๐ 2
,
with ๐›พ๐‘– = 1 โˆ’ ๐œ†๐‘–ฮฃ๐‘–๐‘–.
These formulae are used to find the maximum points ๐€ ๐‘€๐‘ƒ and ๐›ผ ๐‘€๐‘ƒ.
25
1. Starting values ๐›ผ = ๐œŽโˆ’2(๐’•), ๐€ = ๐Ÿ
2. Calculate ฮฃ = ฮ› + ๐›ผฮฆ ๐‘‡ฮฆ โˆ’1 and ๐’˜ = ๐ = ๐›ผฮฃฮฆT ๐ญ
3. Update ๐œ†๐‘– =
๐›พ ๐‘–
๐œ‡ ๐‘–
2 and ๐›ผ =
๐‘โˆ’ ๐‘– ๐›พ ๐‘–
๐’•โˆ’ฮฆ๐ 2 where ๐›พ๐‘– = 1 โˆ’ ๐œ†๐‘–ฮฃ๐‘–๐‘–
4. Prune ๐œ†๐‘– and ๐œ™๐‘– if ๐œ†๐‘– > ๐œ† ๐‘กโ„Ž๐‘Ÿ๐‘’๐‘ โ„Ž๐‘œ๐‘™๐‘‘
5. If not converged go to 2.
Sklearn implementation:
The parameters ๐›ผ1, ๐›ผ2 as well as ๐œ†1, ๐œ†2 are the hyperprior parameters
for ๐›ผ and ๐€ with
๐‘ƒ ๐›ผ โˆผ ฮ“ ๐›ผ1, ๐›ผ2
โˆ’1
, ๐‘ƒ ๐œ†๐‘– โˆผ ฮ“ ๐œ†1, ๐œ†2
โˆ’1
.
๐ธ ฮ“ ๐›ผ, ๐›ฝ =
๐›ผ
๐›ฝ
and ๐‘‰ ฮ“ ๐›ผ, ๐›ฝ =
๐›ผ
๐›ฝ2.
26
Given a some new data ๐‘ฅโˆ—, a prediction for ๐‘กโˆ— is made by
๐‘ƒ ๐‘กโˆ— ๐’•, ๐€ ๐‘€๐‘ƒ, ๐›ผ ๐‘€๐‘ƒ = ๐‘ƒ ๐‘กโˆ— ๐’˜, ๐›ผ ๐‘€๐‘ƒ ๐‘ƒ ๐’˜ ๐’•, ๐€ ๐‘€๐‘ƒ, ๐›ผ ๐‘€๐‘ƒ ๐‘‘๐’˜
= ๐’ฉ ๐ ๐‘‡ ๐œ™ ๐‘ฅโˆ— , ๐›ผ ๐‘€๐‘ƒ
โˆ’1
+ ๐œ™ ๐‘ฅโˆ—
๐‘กฮฃ๐œ™ ๐‘ฅโˆ— .
This is a good approximation of the predictive distribution
๐‘ƒ ๐‘กโˆ— ๐’• = ๐‘ƒ ๐‘กโˆ— ๐’˜, ๐€, ๐›ผ ๐‘ƒ ๐’˜, ๐€, ๐›ผ ๐’• ๐‘‘๐’˜ ๐‘‘๐€ ๐‘‘ฮฑ .
27
1. D. J. C. MacKay. Bayesian Interpolation. 1992
(โ€ฆ to understand the overall idea)
2. M. E.Tipping. Sparse Bayesian learning and the RelevanceVector
Machine. June, 2001
(โ€ฆ to understand the ARD algorithm)
3. T. Fletcher. RelevanceVector Machines Explained. October, 2010
(โ€ฆ to understand the ARD algorithm in detail)
4. D.Wipf. A NewView of Automatic Relevance Determination. 2008
(โ€ฆ not as good as the ones above)
Graphs from slides 7 and 9 were taken from [1] and the awesome
tutorials of Scikit-Learn were consulted many times.
28
29

Weitere รคhnliche Inhalte

Was ist angesagt?

ใƒ™ใ‚คใ‚บ็ตฑ่จˆๅญฆใฎๆฆ‚่ซ–็š„็ดนไป‹
ใƒ™ใ‚คใ‚บ็ตฑ่จˆๅญฆใฎๆฆ‚่ซ–็š„็ดนไป‹ใƒ™ใ‚คใ‚บ็ตฑ่จˆๅญฆใฎๆฆ‚่ซ–็š„็ดนไป‹
ใƒ™ใ‚คใ‚บ็ตฑ่จˆๅญฆใฎๆฆ‚่ซ–็š„็ดนไป‹Naoki Hayashi
ย 
PRMLrevenge_3.3
PRMLrevenge_3.3PRMLrevenge_3.3
PRMLrevenge_3.3Naoya Nakamura
ย 
Outlier detection by Ueda's method
Outlier detection by Ueda's methodOutlier detection by Ueda's method
Outlier detection by Ueda's methodPOOJA PATIL
ย 
Lasso and ridge regression
Lasso and ridge regressionLasso and ridge regression
Lasso and ridge regressionSreerajVA
ย 
Computational learning theory
Computational learning theoryComputational learning theory
Computational learning theoryswapnac12
ย 
Approximate Inference (Chapter 10, PRML Reading)
Approximate Inference (Chapter 10, PRML Reading)Approximate Inference (Chapter 10, PRML Reading)
Approximate Inference (Chapter 10, PRML Reading)Ha Phuong
ย 
ใƒŽใƒณใƒ‘ใƒฉใƒ™ใ‚คใ‚บๅ…ฅ้–€ใฎๅ…ฅ้–€
ใƒŽใƒณใƒ‘ใƒฉใƒ™ใ‚คใ‚บๅ…ฅ้–€ใฎๅ…ฅ้–€ใƒŽใƒณใƒ‘ใƒฉใƒ™ใ‚คใ‚บๅ…ฅ้–€ใฎๅ…ฅ้–€
ใƒŽใƒณใƒ‘ใƒฉใƒ™ใ‚คใ‚บๅ…ฅ้–€ใฎๅ…ฅ้–€Shuyo Nakatani
ย 
ๆฑไบฌ้ƒฝๅธ‚ๅคงๅญฆ ใƒ‡ใƒผใ‚ฟ่งฃๆžๅ…ฅ้–€ 5 ใ‚นใƒ‘ใƒผใ‚นๆ€งใจๅœง็ธฎใ‚ปใƒณใ‚ทใƒณใ‚ฐ 2
ๆฑไบฌ้ƒฝๅธ‚ๅคงๅญฆ ใƒ‡ใƒผใ‚ฟ่งฃๆžๅ…ฅ้–€ 5 ใ‚นใƒ‘ใƒผใ‚นๆ€งใจๅœง็ธฎใ‚ปใƒณใ‚ทใƒณใ‚ฐ 2ๆฑไบฌ้ƒฝๅธ‚ๅคงๅญฆ ใƒ‡ใƒผใ‚ฟ่งฃๆžๅ…ฅ้–€ 5 ใ‚นใƒ‘ใƒผใ‚นๆ€งใจๅœง็ธฎใ‚ปใƒณใ‚ทใƒณใ‚ฐ 2
ๆฑไบฌ้ƒฝๅธ‚ๅคงๅญฆ ใƒ‡ใƒผใ‚ฟ่งฃๆžๅ…ฅ้–€ 5 ใ‚นใƒ‘ใƒผใ‚นๆ€งใจๅœง็ธฎใ‚ปใƒณใ‚ทใƒณใ‚ฐ 2hirokazutanaka
ย 
Rใงใฎtry้–ขๆ•ฐใซใ‚ˆใ‚‹ใ‚จใƒฉใƒผๅ‡ฆ็†
Rใงใฎtry้–ขๆ•ฐใซใ‚ˆใ‚‹ใ‚จใƒฉใƒผๅ‡ฆ็†Rใงใฎtry้–ขๆ•ฐใซใ‚ˆใ‚‹ใ‚จใƒฉใƒผๅ‡ฆ็†
Rใงใฎtry้–ขๆ•ฐใซใ‚ˆใ‚‹ใ‚จใƒฉใƒผๅ‡ฆ็†wada, kazumi
ย 
Bayes rule (Bayes Law)
Bayes rule (Bayes Law)Bayes rule (Bayes Law)
Bayes rule (Bayes Law)Tish997
ย 
ใ‚ฏใƒฉใ‚ทใƒƒใ‚ฏใชๆฉŸๆขฐๅญฆ็ฟ’ใฎๅ…ฅ้–€ 6. ๆœ€้ฉๅŒ–ใจๅญฆ็ฟ’ใ‚ขใƒซใ‚ดใƒชใ‚บใƒ 
ใ‚ฏใƒฉใ‚ทใƒƒใ‚ฏใชๆฉŸๆขฐๅญฆ็ฟ’ใฎๅ…ฅ้–€  6. ๆœ€้ฉๅŒ–ใจๅญฆ็ฟ’ใ‚ขใƒซใ‚ดใƒชใ‚บใƒ ใ‚ฏใƒฉใ‚ทใƒƒใ‚ฏใชๆฉŸๆขฐๅญฆ็ฟ’ใฎๅ…ฅ้–€  6. ๆœ€้ฉๅŒ–ใจๅญฆ็ฟ’ใ‚ขใƒซใ‚ดใƒชใ‚บใƒ 
ใ‚ฏใƒฉใ‚ทใƒƒใ‚ฏใชๆฉŸๆขฐๅญฆ็ฟ’ใฎๅ…ฅ้–€ 6. ๆœ€้ฉๅŒ–ใจๅญฆ็ฟ’ใ‚ขใƒซใ‚ดใƒชใ‚บใƒ Hiroshi Nakagawa
ย 
Fast Unbalanced Optimal Transport on a Tree
Fast Unbalanced Optimal Transport on a TreeFast Unbalanced Optimal Transport on a Tree
Fast Unbalanced Optimal Transport on a Treejoisino
ย 
Interpretable Machine Learning
Interpretable Machine LearningInterpretable Machine Learning
Interpretable Machine LearningSri Ambati
ย 
Gradient descent method
Gradient descent methodGradient descent method
Gradient descent methodSanghyuk Chun
ย 
PRML่ผช่ชญ#10
PRML่ผช่ชญ#10PRML่ผช่ชญ#10
PRML่ผช่ชญ#10matsuolab
ย 
Boosting - An Ensemble Machine Learning Method
Boosting - An Ensemble Machine Learning MethodBoosting - An Ensemble Machine Learning Method
Boosting - An Ensemble Machine Learning MethodKirkwood Donavin
ย 
ๅˆถ้™ใƒœใƒซใƒ„ใƒžใƒณใƒžใ‚ทใƒณๅ…ฅ้–€
ๅˆถ้™ใƒœใƒซใƒ„ใƒžใƒณใƒžใ‚ทใƒณๅ…ฅ้–€ๅˆถ้™ใƒœใƒซใƒ„ใƒžใƒณใƒžใ‚ทใƒณๅ…ฅ้–€
ๅˆถ้™ใƒœใƒซใƒ„ใƒžใƒณใƒžใ‚ทใƒณๅ…ฅ้–€ไฝ‘้ฆฌ ๆ–Ž่—ค
ย 
InfoGAN: Interpretable Representation Learning by Information Maximizing Gene...
InfoGAN: Interpretable Representation Learning by Information Maximizing Gene...InfoGAN: Interpretable Representation Learning by Information Maximizing Gene...
InfoGAN: Interpretable Representation Learning by Information Maximizing Gene...Joonhyung Lee
ย 
PRML่ผช่ชญ#11
PRML่ผช่ชญ#11PRML่ผช่ชญ#11
PRML่ผช่ชญ#11matsuolab
ย 

Was ist angesagt? (20)

ใƒ™ใ‚คใ‚บ็ตฑ่จˆๅญฆใฎๆฆ‚่ซ–็š„็ดนไป‹
ใƒ™ใ‚คใ‚บ็ตฑ่จˆๅญฆใฎๆฆ‚่ซ–็š„็ดนไป‹ใƒ™ใ‚คใ‚บ็ตฑ่จˆๅญฆใฎๆฆ‚่ซ–็š„็ดนไป‹
ใƒ™ใ‚คใ‚บ็ตฑ่จˆๅญฆใฎๆฆ‚่ซ–็š„็ดนไป‹
ย 
PRMLrevenge_3.3
PRMLrevenge_3.3PRMLrevenge_3.3
PRMLrevenge_3.3
ย 
Outlier detection by Ueda's method
Outlier detection by Ueda's methodOutlier detection by Ueda's method
Outlier detection by Ueda's method
ย 
Lasso and ridge regression
Lasso and ridge regressionLasso and ridge regression
Lasso and ridge regression
ย 
Computational learning theory
Computational learning theoryComputational learning theory
Computational learning theory
ย 
Approximate Inference (Chapter 10, PRML Reading)
Approximate Inference (Chapter 10, PRML Reading)Approximate Inference (Chapter 10, PRML Reading)
Approximate Inference (Chapter 10, PRML Reading)
ย 
ใƒŽใƒณใƒ‘ใƒฉใƒ™ใ‚คใ‚บๅ…ฅ้–€ใฎๅ…ฅ้–€
ใƒŽใƒณใƒ‘ใƒฉใƒ™ใ‚คใ‚บๅ…ฅ้–€ใฎๅ…ฅ้–€ใƒŽใƒณใƒ‘ใƒฉใƒ™ใ‚คใ‚บๅ…ฅ้–€ใฎๅ…ฅ้–€
ใƒŽใƒณใƒ‘ใƒฉใƒ™ใ‚คใ‚บๅ…ฅ้–€ใฎๅ…ฅ้–€
ย 
ๆฑไบฌ้ƒฝๅธ‚ๅคงๅญฆ ใƒ‡ใƒผใ‚ฟ่งฃๆžๅ…ฅ้–€ 5 ใ‚นใƒ‘ใƒผใ‚นๆ€งใจๅœง็ธฎใ‚ปใƒณใ‚ทใƒณใ‚ฐ 2
ๆฑไบฌ้ƒฝๅธ‚ๅคงๅญฆ ใƒ‡ใƒผใ‚ฟ่งฃๆžๅ…ฅ้–€ 5 ใ‚นใƒ‘ใƒผใ‚นๆ€งใจๅœง็ธฎใ‚ปใƒณใ‚ทใƒณใ‚ฐ 2ๆฑไบฌ้ƒฝๅธ‚ๅคงๅญฆ ใƒ‡ใƒผใ‚ฟ่งฃๆžๅ…ฅ้–€ 5 ใ‚นใƒ‘ใƒผใ‚นๆ€งใจๅœง็ธฎใ‚ปใƒณใ‚ทใƒณใ‚ฐ 2
ๆฑไบฌ้ƒฝๅธ‚ๅคงๅญฆ ใƒ‡ใƒผใ‚ฟ่งฃๆžๅ…ฅ้–€ 5 ใ‚นใƒ‘ใƒผใ‚นๆ€งใจๅœง็ธฎใ‚ปใƒณใ‚ทใƒณใ‚ฐ 2
ย 
Bayesian learning
Bayesian learningBayesian learning
Bayesian learning
ย 
Rใงใฎtry้–ขๆ•ฐใซใ‚ˆใ‚‹ใ‚จใƒฉใƒผๅ‡ฆ็†
Rใงใฎtry้–ขๆ•ฐใซใ‚ˆใ‚‹ใ‚จใƒฉใƒผๅ‡ฆ็†Rใงใฎtry้–ขๆ•ฐใซใ‚ˆใ‚‹ใ‚จใƒฉใƒผๅ‡ฆ็†
Rใงใฎtry้–ขๆ•ฐใซใ‚ˆใ‚‹ใ‚จใƒฉใƒผๅ‡ฆ็†
ย 
Bayes rule (Bayes Law)
Bayes rule (Bayes Law)Bayes rule (Bayes Law)
Bayes rule (Bayes Law)
ย 
ใ‚ฏใƒฉใ‚ทใƒƒใ‚ฏใชๆฉŸๆขฐๅญฆ็ฟ’ใฎๅ…ฅ้–€ 6. ๆœ€้ฉๅŒ–ใจๅญฆ็ฟ’ใ‚ขใƒซใ‚ดใƒชใ‚บใƒ 
ใ‚ฏใƒฉใ‚ทใƒƒใ‚ฏใชๆฉŸๆขฐๅญฆ็ฟ’ใฎๅ…ฅ้–€  6. ๆœ€้ฉๅŒ–ใจๅญฆ็ฟ’ใ‚ขใƒซใ‚ดใƒชใ‚บใƒ ใ‚ฏใƒฉใ‚ทใƒƒใ‚ฏใชๆฉŸๆขฐๅญฆ็ฟ’ใฎๅ…ฅ้–€  6. ๆœ€้ฉๅŒ–ใจๅญฆ็ฟ’ใ‚ขใƒซใ‚ดใƒชใ‚บใƒ 
ใ‚ฏใƒฉใ‚ทใƒƒใ‚ฏใชๆฉŸๆขฐๅญฆ็ฟ’ใฎๅ…ฅ้–€ 6. ๆœ€้ฉๅŒ–ใจๅญฆ็ฟ’ใ‚ขใƒซใ‚ดใƒชใ‚บใƒ 
ย 
Fast Unbalanced Optimal Transport on a Tree
Fast Unbalanced Optimal Transport on a TreeFast Unbalanced Optimal Transport on a Tree
Fast Unbalanced Optimal Transport on a Tree
ย 
Interpretable Machine Learning
Interpretable Machine LearningInterpretable Machine Learning
Interpretable Machine Learning
ย 
Gradient descent method
Gradient descent methodGradient descent method
Gradient descent method
ย 
PRML่ผช่ชญ#10
PRML่ผช่ชญ#10PRML่ผช่ชญ#10
PRML่ผช่ชญ#10
ย 
Boosting - An Ensemble Machine Learning Method
Boosting - An Ensemble Machine Learning MethodBoosting - An Ensemble Machine Learning Method
Boosting - An Ensemble Machine Learning Method
ย 
ๅˆถ้™ใƒœใƒซใƒ„ใƒžใƒณใƒžใ‚ทใƒณๅ…ฅ้–€
ๅˆถ้™ใƒœใƒซใƒ„ใƒžใƒณใƒžใ‚ทใƒณๅ…ฅ้–€ๅˆถ้™ใƒœใƒซใƒ„ใƒžใƒณใƒžใ‚ทใƒณๅ…ฅ้–€
ๅˆถ้™ใƒœใƒซใƒ„ใƒžใƒณใƒžใ‚ทใƒณๅ…ฅ้–€
ย 
InfoGAN: Interpretable Representation Learning by Information Maximizing Gene...
InfoGAN: Interpretable Representation Learning by Information Maximizing Gene...InfoGAN: Interpretable Representation Learning by Information Maximizing Gene...
InfoGAN: Interpretable Representation Learning by Information Maximizing Gene...
ย 
PRML่ผช่ชญ#11
PRML่ผช่ชญ#11PRML่ผช่ชญ#11
PRML่ผช่ชญ#11
ย 

Andere mochten auch

t4_20110728_IGARSS11_tsutomuy.pdf
t4_20110728_IGARSS11_tsutomuy.pdft4_20110728_IGARSS11_tsutomuy.pdf
t4_20110728_IGARSS11_tsutomuy.pdfgrssieee
ย 
Prigogine, i. tan solo una_ilusion
Prigogine, i. tan solo una_ilusionPrigogine, i. tan solo una_ilusion
Prigogine, i. tan solo una_ilusionJoan M Ferrer Dรญaz
ย 
Inference for stochastic differential equations via approximate Bayesian comp...
Inference for stochastic differential equations via approximate Bayesian comp...Inference for stochastic differential equations via approximate Bayesian comp...
Inference for stochastic differential equations via approximate Bayesian comp...Umberto Picchini
ย 
Hirsch s.w., smale s. differential equations, dynamical systems and linear ...
Hirsch s.w., smale s.   differential equations, dynamical systems and linear ...Hirsch s.w., smale s.   differential equations, dynamical systems and linear ...
Hirsch s.w., smale s. differential equations, dynamical systems and linear ...yuejia2001
ย 
Libro de Mancil
Libro de MancilLibro de Mancil
Libro de MancilLenore Corella
ย 
Un Juego Diferencial Estocรกstico para Reaseguro
Un Juego Diferencial Estocรกstico para ReaseguroUn Juego Diferencial Estocรกstico para Reaseguro
Un Juego Diferencial Estocรกstico para ReaseguroDavid Solis
ย 
MODELACIร“N MATEMรTICA A TRAVร‰S DE LAS ECUACIONES EN DIFERENCIA
MODELACIร“N MATEMรTICA A TRAVร‰S DE LAS ECUACIONES EN DIFERENCIAMODELACIร“N MATEMรTICA A TRAVร‰S DE LAS ECUACIONES EN DIFERENCIA
MODELACIร“N MATEMรTICA A TRAVร‰S DE LAS ECUACIONES EN DIFERENCIANubia Mejรญa
ย 
Financial Markets with Stochastic Volatilities - markov modelling
Financial Markets with Stochastic Volatilities - markov modellingFinancial Markets with Stochastic Volatilities - markov modelling
Financial Markets with Stochastic Volatilities - markov modellingguest8901f4
ย 
Calculo diferencial e_integral_en_la_vida_cotidiana (2)
Calculo diferencial e_integral_en_la_vida_cotidiana (2)Calculo diferencial e_integral_en_la_vida_cotidiana (2)
Calculo diferencial e_integral_en_la_vida_cotidiana (2)Hugo Rosales Vera
ย 
Calculo diferencial e integral
Calculo diferencial e integralCalculo diferencial e integral
Calculo diferencial e integralronalrepi1
ย 
Calculo integral
Calculo integralCalculo integral
Calculo integralkarlaaflores
ย 
Solucionario de matematicas de g. mancill.
Solucionario de matematicas de g. mancill.Solucionario de matematicas de g. mancill.
Solucionario de matematicas de g. mancill.C 12
ย 
Fractales y Teorรญa del Caos
Fractales y Teorรญa del CaosFractales y Teorรญa del Caos
Fractales y Teorรญa del CaosJimmy Campo
ย 
Ejercicios algebra superior hall y knight
Ejercicios algebra superior hall y knightEjercicios algebra superior hall y knight
Ejercicios algebra superior hall y knightguido guzman perez
ย 
Algebra Elemental Moderna
Algebra Elemental ModernaAlgebra Elemental Moderna
Algebra Elemental ModernaRicardo Avila Legra
ย 
Algebra proschle
Algebra  proschleAlgebra  proschle
Algebra proschle3DSalesianos
ย 
Algebra arrayan
Algebra arrayanAlgebra arrayan
Algebra arrayangomezaa
ย 

Andere mochten auch (20)

t4_20110728_IGARSS11_tsutomuy.pdf
t4_20110728_IGARSS11_tsutomuy.pdft4_20110728_IGARSS11_tsutomuy.pdf
t4_20110728_IGARSS11_tsutomuy.pdf
ย 
Prigogine, i. tan solo una_ilusion
Prigogine, i. tan solo una_ilusionPrigogine, i. tan solo una_ilusion
Prigogine, i. tan solo una_ilusion
ย 
Inference for stochastic differential equations via approximate Bayesian comp...
Inference for stochastic differential equations via approximate Bayesian comp...Inference for stochastic differential equations via approximate Bayesian comp...
Inference for stochastic differential equations via approximate Bayesian comp...
ย 
Hirsch s.w., smale s. differential equations, dynamical systems and linear ...
Hirsch s.w., smale s.   differential equations, dynamical systems and linear ...Hirsch s.w., smale s.   differential equations, dynamical systems and linear ...
Hirsch s.w., smale s. differential equations, dynamical systems and linear ...
ย 
Francisco
FranciscoFrancisco
Francisco
ย 
Libro de Mancil
Libro de MancilLibro de Mancil
Libro de Mancil
ย 
Un Juego Diferencial Estocรกstico para Reaseguro
Un Juego Diferencial Estocรกstico para ReaseguroUn Juego Diferencial Estocรกstico para Reaseguro
Un Juego Diferencial Estocรกstico para Reaseguro
ย 
Prigogine esayo
Prigogine esayoPrigogine esayo
Prigogine esayo
ย 
Calculo diferencial e integral2
Calculo diferencial e integral2Calculo diferencial e integral2
Calculo diferencial e integral2
ย 
MODELACIร“N MATEMรTICA A TRAVร‰S DE LAS ECUACIONES EN DIFERENCIA
MODELACIร“N MATEMรTICA A TRAVร‰S DE LAS ECUACIONES EN DIFERENCIAMODELACIร“N MATEMรTICA A TRAVร‰S DE LAS ECUACIONES EN DIFERENCIA
MODELACIร“N MATEMรTICA A TRAVร‰S DE LAS ECUACIONES EN DIFERENCIA
ย 
Financial Markets with Stochastic Volatilities - markov modelling
Financial Markets with Stochastic Volatilities - markov modellingFinancial Markets with Stochastic Volatilities - markov modelling
Financial Markets with Stochastic Volatilities - markov modelling
ย 
Calculo diferencial e_integral_en_la_vida_cotidiana (2)
Calculo diferencial e_integral_en_la_vida_cotidiana (2)Calculo diferencial e_integral_en_la_vida_cotidiana (2)
Calculo diferencial e_integral_en_la_vida_cotidiana (2)
ย 
Calculo diferencial e integral
Calculo diferencial e integralCalculo diferencial e integral
Calculo diferencial e integral
ย 
Calculo integral
Calculo integralCalculo integral
Calculo integral
ย 
Solucionario de matematicas de g. mancill.
Solucionario de matematicas de g. mancill.Solucionario de matematicas de g. mancill.
Solucionario de matematicas de g. mancill.
ย 
Fractales y Teorรญa del Caos
Fractales y Teorรญa del CaosFractales y Teorรญa del Caos
Fractales y Teorรญa del Caos
ย 
Ejercicios algebra superior hall y knight
Ejercicios algebra superior hall y knightEjercicios algebra superior hall y knight
Ejercicios algebra superior hall y knight
ย 
Algebra Elemental Moderna
Algebra Elemental ModernaAlgebra Elemental Moderna
Algebra Elemental Moderna
ย 
Algebra proschle
Algebra  proschleAlgebra  proschle
Algebra proschle
ย 
Algebra arrayan
Algebra arrayanAlgebra arrayan
Algebra arrayan
ย 

ร„hnlich wie Explaining the idea behind automatic relevance determination and bayesian interpolation

Koh_Liang_ICML2017
Koh_Liang_ICML2017Koh_Liang_ICML2017
Koh_Liang_ICML2017Masa Kato
ย 
Robot, Learning From Data
Robot, Learning From DataRobot, Learning From Data
Robot, Learning From DataSungjoon Choi
ย 
Generalised Statistical Convergence For Double Sequences
Generalised Statistical Convergence For Double SequencesGeneralised Statistical Convergence For Double Sequences
Generalised Statistical Convergence For Double SequencesIOSR Journals
ย 
Boundness of a neural network weights using the notion of a limit of a sequence
Boundness of a neural network weights using the notion of a limit of a sequenceBoundness of a neural network weights using the notion of a limit of a sequence
Boundness of a neural network weights using the notion of a limit of a sequenceIJDKP
ย 
STLtalk about statistical analysis and its application
STLtalk about statistical analysis and its applicationSTLtalk about statistical analysis and its application
STLtalk about statistical analysis and its applicationJulieDash5
ย 
MM - KBAC: Using mixed models to adjust for population structure in a rare-va...
MM - KBAC: Using mixed models to adjust for population structure in a rare-va...MM - KBAC: Using mixed models to adjust for population structure in a rare-va...
MM - KBAC: Using mixed models to adjust for population structure in a rare-va...Golden Helix Inc
ย 
Machine Learning 1
Machine Learning 1Machine Learning 1
Machine Learning 1cairo university
ย 
Mncs 16-09-4์ฃผ-๋ณ€์Šน๊ทœ-introduction to the machine learning
Mncs 16-09-4์ฃผ-๋ณ€์Šน๊ทœ-introduction to the machine learningMncs 16-09-4์ฃผ-๋ณ€์Šน๊ทœ-introduction to the machine learning
Mncs 16-09-4์ฃผ-๋ณ€์Šน๊ทœ-introduction to the machine learningSeung-gyu Byeon
ย 
Symbolic Computation via Grรถbner Basis
Symbolic Computation via Grรถbner BasisSymbolic Computation via Grรถbner Basis
Symbolic Computation via Grรถbner BasisIJERA Editor
ย 
Intro to Quant Trading Strategies (Lecture 2 of 10)
Intro to Quant Trading Strategies (Lecture 2 of 10)Intro to Quant Trading Strategies (Lecture 2 of 10)
Intro to Quant Trading Strategies (Lecture 2 of 10)Adrian Aley
ย 
A machine learning method for efficient design optimization in nano-optics
A machine learning method for efficient design optimization in nano-optics A machine learning method for efficient design optimization in nano-optics
A machine learning method for efficient design optimization in nano-optics JCMwave
ย 
Anti-differentiating Approximation Algorithms: PageRank and MinCut
Anti-differentiating Approximation Algorithms: PageRank and MinCutAnti-differentiating Approximation Algorithms: PageRank and MinCut
Anti-differentiating Approximation Algorithms: PageRank and MinCutDavid Gleich
ย 
Machine learning ppt and presentation code
Machine learning ppt and presentation codeMachine learning ppt and presentation code
Machine learning ppt and presentation codesharma239172
ย 
Learning group em - 20171025 - copy
Learning group   em - 20171025 - copyLearning group   em - 20171025 - copy
Learning group em - 20171025 - copyShuai Zhang
ย 
07 Machine Learning - Expectation Maximization
07 Machine Learning - Expectation Maximization07 Machine Learning - Expectation Maximization
07 Machine Learning - Expectation MaximizationAndres Mendez-Vazquez
ย 
Ck31369376
Ck31369376Ck31369376
Ck31369376IJMER
ย 
Kernel Bayes Rule
Kernel Bayes RuleKernel Bayes Rule
Kernel Bayes RuleYan Xu
ย 
PRML Chapter 6
PRML Chapter 6PRML Chapter 6
PRML Chapter 6Sunwoo Kim
ย 
A PROBABILISTIC ALGORITHM OF COMPUTING THE POLYNOMIAL GREATEST COMMON DIVISOR...
A PROBABILISTIC ALGORITHM OF COMPUTING THE POLYNOMIAL GREATEST COMMON DIVISOR...A PROBABILISTIC ALGORITHM OF COMPUTING THE POLYNOMIAL GREATEST COMMON DIVISOR...
A PROBABILISTIC ALGORITHM OF COMPUTING THE POLYNOMIAL GREATEST COMMON DIVISOR...ijscmcj
ย 
PRML Chapter 8
PRML Chapter 8PRML Chapter 8
PRML Chapter 8Sunwoo Kim
ย 

ร„hnlich wie Explaining the idea behind automatic relevance determination and bayesian interpolation (20)

Koh_Liang_ICML2017
Koh_Liang_ICML2017Koh_Liang_ICML2017
Koh_Liang_ICML2017
ย 
Robot, Learning From Data
Robot, Learning From DataRobot, Learning From Data
Robot, Learning From Data
ย 
Generalised Statistical Convergence For Double Sequences
Generalised Statistical Convergence For Double SequencesGeneralised Statistical Convergence For Double Sequences
Generalised Statistical Convergence For Double Sequences
ย 
Boundness of a neural network weights using the notion of a limit of a sequence
Boundness of a neural network weights using the notion of a limit of a sequenceBoundness of a neural network weights using the notion of a limit of a sequence
Boundness of a neural network weights using the notion of a limit of a sequence
ย 
STLtalk about statistical analysis and its application
STLtalk about statistical analysis and its applicationSTLtalk about statistical analysis and its application
STLtalk about statistical analysis and its application
ย 
MM - KBAC: Using mixed models to adjust for population structure in a rare-va...
MM - KBAC: Using mixed models to adjust for population structure in a rare-va...MM - KBAC: Using mixed models to adjust for population structure in a rare-va...
MM - KBAC: Using mixed models to adjust for population structure in a rare-va...
ย 
Machine Learning 1
Machine Learning 1Machine Learning 1
Machine Learning 1
ย 
Mncs 16-09-4์ฃผ-๋ณ€์Šน๊ทœ-introduction to the machine learning
Mncs 16-09-4์ฃผ-๋ณ€์Šน๊ทœ-introduction to the machine learningMncs 16-09-4์ฃผ-๋ณ€์Šน๊ทœ-introduction to the machine learning
Mncs 16-09-4์ฃผ-๋ณ€์Šน๊ทœ-introduction to the machine learning
ย 
Symbolic Computation via Grรถbner Basis
Symbolic Computation via Grรถbner BasisSymbolic Computation via Grรถbner Basis
Symbolic Computation via Grรถbner Basis
ย 
Intro to Quant Trading Strategies (Lecture 2 of 10)
Intro to Quant Trading Strategies (Lecture 2 of 10)Intro to Quant Trading Strategies (Lecture 2 of 10)
Intro to Quant Trading Strategies (Lecture 2 of 10)
ย 
A machine learning method for efficient design optimization in nano-optics
A machine learning method for efficient design optimization in nano-optics A machine learning method for efficient design optimization in nano-optics
A machine learning method for efficient design optimization in nano-optics
ย 
Anti-differentiating Approximation Algorithms: PageRank and MinCut
Anti-differentiating Approximation Algorithms: PageRank and MinCutAnti-differentiating Approximation Algorithms: PageRank and MinCut
Anti-differentiating Approximation Algorithms: PageRank and MinCut
ย 
Machine learning ppt and presentation code
Machine learning ppt and presentation codeMachine learning ppt and presentation code
Machine learning ppt and presentation code
ย 
Learning group em - 20171025 - copy
Learning group   em - 20171025 - copyLearning group   em - 20171025 - copy
Learning group em - 20171025 - copy
ย 
07 Machine Learning - Expectation Maximization
07 Machine Learning - Expectation Maximization07 Machine Learning - Expectation Maximization
07 Machine Learning - Expectation Maximization
ย 
Ck31369376
Ck31369376Ck31369376
Ck31369376
ย 
Kernel Bayes Rule
Kernel Bayes RuleKernel Bayes Rule
Kernel Bayes Rule
ย 
PRML Chapter 6
PRML Chapter 6PRML Chapter 6
PRML Chapter 6
ย 
A PROBABILISTIC ALGORITHM OF COMPUTING THE POLYNOMIAL GREATEST COMMON DIVISOR...
A PROBABILISTIC ALGORITHM OF COMPUTING THE POLYNOMIAL GREATEST COMMON DIVISOR...A PROBABILISTIC ALGORITHM OF COMPUTING THE POLYNOMIAL GREATEST COMMON DIVISOR...
A PROBABILISTIC ALGORITHM OF COMPUTING THE POLYNOMIAL GREATEST COMMON DIVISOR...
ย 
PRML Chapter 8
PRML Chapter 8PRML Chapter 8
PRML Chapter 8
ย 

Mehr von Florian Wilhelm

Streamlining Python Development: A Guide to a Modern Project Setup
Streamlining Python Development: A Guide to a Modern Project SetupStreamlining Python Development: A Guide to a Modern Project Setup
Streamlining Python Development: A Guide to a Modern Project SetupFlorian Wilhelm
ย 
Unlocking the Power of Integer Programming
Unlocking the Power of Integer ProgrammingUnlocking the Power of Integer Programming
Unlocking the Power of Integer ProgrammingFlorian Wilhelm
ย 
WALD: A Modern & Sustainable Analytics Stack
WALD: A Modern & Sustainable Analytics StackWALD: A Modern & Sustainable Analytics Stack
WALD: A Modern & Sustainable Analytics StackFlorian Wilhelm
ย 
Forget about AI and do Mathematical Modelling instead!
Forget about AI and do Mathematical Modelling instead!Forget about AI and do Mathematical Modelling instead!
Forget about AI and do Mathematical Modelling instead!Florian Wilhelm
ย 
An Interpretable Model for Collaborative Filtering Using an Extended Latent D...
An Interpretable Model for Collaborative Filtering Using an Extended Latent D...An Interpretable Model for Collaborative Filtering Using an Extended Latent D...
An Interpretable Model for Collaborative Filtering Using an Extended Latent D...Florian Wilhelm
ย 
Honey I Shrunk the Target Variable! Common pitfalls when transforming the tar...
Honey I Shrunk the Target Variable! Common pitfalls when transforming the tar...Honey I Shrunk the Target Variable! Common pitfalls when transforming the tar...
Honey I Shrunk the Target Variable! Common pitfalls when transforming the tar...Florian Wilhelm
ย 
Matrix Factorization for Collaborative Filtering Is Just Solving an Adjoint L...
Matrix Factorization for Collaborative Filtering Is Just Solving an Adjoint L...Matrix Factorization for Collaborative Filtering Is Just Solving an Adjoint L...
Matrix Factorization for Collaborative Filtering Is Just Solving an Adjoint L...Florian Wilhelm
ย 
Uncertainty Quantification in AI
Uncertainty Quantification in AIUncertainty Quantification in AI
Uncertainty Quantification in AIFlorian Wilhelm
ย 
Performance evaluation of GANs in a semisupervised OCR use case
Performance evaluation of GANs in a semisupervised OCR use casePerformance evaluation of GANs in a semisupervised OCR use case
Performance evaluation of GANs in a semisupervised OCR use caseFlorian Wilhelm
ย 
Bridging the Gap: from Data Science to Production
Bridging the Gap: from Data Science to ProductionBridging the Gap: from Data Science to Production
Bridging the Gap: from Data Science to ProductionFlorian Wilhelm
ย 
How mobile.de brings Data Science to Production for a Personalized Web Experi...
How mobile.de brings Data Science to Production for a Personalized Web Experi...How mobile.de brings Data Science to Production for a Personalized Web Experi...
How mobile.de brings Data Science to Production for a Personalized Web Experi...Florian Wilhelm
ย 
Deep Learning-based Recommendations for Germany's Biggest Vehicle Marketplace
Deep Learning-based Recommendations for Germany's Biggest Vehicle MarketplaceDeep Learning-based Recommendations for Germany's Biggest Vehicle Marketplace
Deep Learning-based Recommendations for Germany's Biggest Vehicle MarketplaceFlorian Wilhelm
ย 
Deep Learning-based Recommendations for Germany's Biggest Online Vehicle Mark...
Deep Learning-based Recommendations for Germany's Biggest Online Vehicle Mark...Deep Learning-based Recommendations for Germany's Biggest Online Vehicle Mark...
Deep Learning-based Recommendations for Germany's Biggest Online Vehicle Mark...Florian Wilhelm
ย 
Declarative Thinking and Programming
Declarative Thinking and ProgrammingDeclarative Thinking and Programming
Declarative Thinking and ProgrammingFlorian Wilhelm
ย 
Which car fits my life? - PyData Berlin 2017
Which car fits my life? - PyData Berlin 2017Which car fits my life? - PyData Berlin 2017
Which car fits my life? - PyData Berlin 2017Florian Wilhelm
ย 
PyData Meetup Berlin 2017-04-19
PyData Meetup Berlin 2017-04-19PyData Meetup Berlin 2017-04-19
PyData Meetup Berlin 2017-04-19Florian Wilhelm
ย 

Mehr von Florian Wilhelm (16)

Streamlining Python Development: A Guide to a Modern Project Setup
Streamlining Python Development: A Guide to a Modern Project SetupStreamlining Python Development: A Guide to a Modern Project Setup
Streamlining Python Development: A Guide to a Modern Project Setup
ย 
Unlocking the Power of Integer Programming
Unlocking the Power of Integer ProgrammingUnlocking the Power of Integer Programming
Unlocking the Power of Integer Programming
ย 
WALD: A Modern & Sustainable Analytics Stack
WALD: A Modern & Sustainable Analytics StackWALD: A Modern & Sustainable Analytics Stack
WALD: A Modern & Sustainable Analytics Stack
ย 
Forget about AI and do Mathematical Modelling instead!
Forget about AI and do Mathematical Modelling instead!Forget about AI and do Mathematical Modelling instead!
Forget about AI and do Mathematical Modelling instead!
ย 
An Interpretable Model for Collaborative Filtering Using an Extended Latent D...
An Interpretable Model for Collaborative Filtering Using an Extended Latent D...An Interpretable Model for Collaborative Filtering Using an Extended Latent D...
An Interpretable Model for Collaborative Filtering Using an Extended Latent D...
ย 
Honey I Shrunk the Target Variable! Common pitfalls when transforming the tar...
Honey I Shrunk the Target Variable! Common pitfalls when transforming the tar...Honey I Shrunk the Target Variable! Common pitfalls when transforming the tar...
Honey I Shrunk the Target Variable! Common pitfalls when transforming the tar...
ย 
Matrix Factorization for Collaborative Filtering Is Just Solving an Adjoint L...
Matrix Factorization for Collaborative Filtering Is Just Solving an Adjoint L...Matrix Factorization for Collaborative Filtering Is Just Solving an Adjoint L...
Matrix Factorization for Collaborative Filtering Is Just Solving an Adjoint L...
ย 
Uncertainty Quantification in AI
Uncertainty Quantification in AIUncertainty Quantification in AI
Uncertainty Quantification in AI
ย 
Performance evaluation of GANs in a semisupervised OCR use case
Performance evaluation of GANs in a semisupervised OCR use casePerformance evaluation of GANs in a semisupervised OCR use case
Performance evaluation of GANs in a semisupervised OCR use case
ย 
Bridging the Gap: from Data Science to Production
Bridging the Gap: from Data Science to ProductionBridging the Gap: from Data Science to Production
Bridging the Gap: from Data Science to Production
ย 
How mobile.de brings Data Science to Production for a Personalized Web Experi...
How mobile.de brings Data Science to Production for a Personalized Web Experi...How mobile.de brings Data Science to Production for a Personalized Web Experi...
How mobile.de brings Data Science to Production for a Personalized Web Experi...
ย 
Deep Learning-based Recommendations for Germany's Biggest Vehicle Marketplace
Deep Learning-based Recommendations for Germany's Biggest Vehicle MarketplaceDeep Learning-based Recommendations for Germany's Biggest Vehicle Marketplace
Deep Learning-based Recommendations for Germany's Biggest Vehicle Marketplace
ย 
Deep Learning-based Recommendations for Germany's Biggest Online Vehicle Mark...
Deep Learning-based Recommendations for Germany's Biggest Online Vehicle Mark...Deep Learning-based Recommendations for Germany's Biggest Online Vehicle Mark...
Deep Learning-based Recommendations for Germany's Biggest Online Vehicle Mark...
ย 
Declarative Thinking and Programming
Declarative Thinking and ProgrammingDeclarative Thinking and Programming
Declarative Thinking and Programming
ย 
Which car fits my life? - PyData Berlin 2017
Which car fits my life? - PyData Berlin 2017Which car fits my life? - PyData Berlin 2017
Which car fits my life? - PyData Berlin 2017
ย 
PyData Meetup Berlin 2017-04-19
PyData Meetup Berlin 2017-04-19PyData Meetup Berlin 2017-04-19
PyData Meetup Berlin 2017-04-19
ย 

Kรผrzlich hochgeladen

VidaXL dropshipping via API with DroFx.pptx
VidaXL dropshipping via API with DroFx.pptxVidaXL dropshipping via API with DroFx.pptx
VidaXL dropshipping via API with DroFx.pptxolyaivanovalion
ย 
Beautiful Sapna Vip Call Girls Hauz Khas 9711199012 Call /Whatsapps
Beautiful Sapna Vip  Call Girls Hauz Khas 9711199012 Call /WhatsappsBeautiful Sapna Vip  Call Girls Hauz Khas 9711199012 Call /Whatsapps
Beautiful Sapna Vip Call Girls Hauz Khas 9711199012 Call /Whatsappssapnasaifi408
ย 
Customer Service Analytics - Make Sense of All Your Data.pptx
Customer Service Analytics - Make Sense of All Your Data.pptxCustomer Service Analytics - Make Sense of All Your Data.pptx
Customer Service Analytics - Make Sense of All Your Data.pptxEmmanuel Dauda
ย 
VIP High Class Call Girls Jamshedpur Anushka 8250192130 Independent Escort Se...
VIP High Class Call Girls Jamshedpur Anushka 8250192130 Independent Escort Se...VIP High Class Call Girls Jamshedpur Anushka 8250192130 Independent Escort Se...
VIP High Class Call Girls Jamshedpur Anushka 8250192130 Independent Escort Se...Suhani Kapoor
ย 
VIP Call Girls Service Miyapur Hyderabad Call +91-8250192130
VIP Call Girls Service Miyapur Hyderabad Call +91-8250192130VIP Call Girls Service Miyapur Hyderabad Call +91-8250192130
VIP Call Girls Service Miyapur Hyderabad Call +91-8250192130Suhani Kapoor
ย 
Schema on read is obsolete. Welcome metaprogramming..pdf
Schema on read is obsolete. Welcome metaprogramming..pdfSchema on read is obsolete. Welcome metaprogramming..pdf
Schema on read is obsolete. Welcome metaprogramming..pdfLars Albertsson
ย 
CebaBaby dropshipping via API with DroFX.pptx
CebaBaby dropshipping via API with DroFX.pptxCebaBaby dropshipping via API with DroFX.pptx
CebaBaby dropshipping via API with DroFX.pptxolyaivanovalion
ย 
April 2024 - Crypto Market Report's Analysis
April 2024 - Crypto Market Report's AnalysisApril 2024 - Crypto Market Report's Analysis
April 2024 - Crypto Market Report's Analysismanisha194592
ย 
Delhi Call Girls Punjabi Bagh 9711199171 โ˜Žโœ”๐Ÿ‘Œโœ” Whatsapp Hard And Sexy Vip Call
Delhi Call Girls Punjabi Bagh 9711199171 โ˜Žโœ”๐Ÿ‘Œโœ” Whatsapp Hard And Sexy Vip CallDelhi Call Girls Punjabi Bagh 9711199171 โ˜Žโœ”๐Ÿ‘Œโœ” Whatsapp Hard And Sexy Vip Call
Delhi Call Girls Punjabi Bagh 9711199171 โ˜Žโœ”๐Ÿ‘Œโœ” Whatsapp Hard And Sexy Vip Callshivangimorya083
ย 
100-Concepts-of-AI by Anupama Kate .pptx
100-Concepts-of-AI by Anupama Kate .pptx100-Concepts-of-AI by Anupama Kate .pptx
100-Concepts-of-AI by Anupama Kate .pptxAnupama Kate
ย 
Invezz.com - Grow your wealth with trading signals
Invezz.com - Grow your wealth with trading signalsInvezz.com - Grow your wealth with trading signals
Invezz.com - Grow your wealth with trading signalsInvezz1
ย 
Log Analysis using OSSEC sasoasasasas.pptx
Log Analysis using OSSEC sasoasasasas.pptxLog Analysis using OSSEC sasoasasasas.pptx
Log Analysis using OSSEC sasoasasasas.pptxJohnnyPlasten
ย 
Smarteg dropshipping via API with DroFx.pptx
Smarteg dropshipping via API with DroFx.pptxSmarteg dropshipping via API with DroFx.pptx
Smarteg dropshipping via API with DroFx.pptxolyaivanovalion
ย 
Midocean dropshipping via API with DroFx
Midocean dropshipping via API with DroFxMidocean dropshipping via API with DroFx
Midocean dropshipping via API with DroFxolyaivanovalion
ย 
Delhi Call Girls CP 9711199171 โ˜Žโœ”๐Ÿ‘Œโœ” Whatsapp Hard And Sexy Vip Call
Delhi Call Girls CP 9711199171 โ˜Žโœ”๐Ÿ‘Œโœ” Whatsapp Hard And Sexy Vip CallDelhi Call Girls CP 9711199171 โ˜Žโœ”๐Ÿ‘Œโœ” Whatsapp Hard And Sexy Vip Call
Delhi Call Girls CP 9711199171 โ˜Žโœ”๐Ÿ‘Œโœ” Whatsapp Hard And Sexy Vip Callshivangimorya083
ย 
Unveiling Insights: The Role of a Data Analyst
Unveiling Insights: The Role of a Data AnalystUnveiling Insights: The Role of a Data Analyst
Unveiling Insights: The Role of a Data AnalystSamantha Rae Coolbeth
ย 
VIP High Profile Call Girls Amravati Aarushi 8250192130 Independent Escort Se...
VIP High Profile Call Girls Amravati Aarushi 8250192130 Independent Escort Se...VIP High Profile Call Girls Amravati Aarushi 8250192130 Independent Escort Se...
VIP High Profile Call Girls Amravati Aarushi 8250192130 Independent Escort Se...Suhani Kapoor
ย 
Market Analysis in the 5 Largest Economic Countries in Southeast Asia.pdf
Market Analysis in the 5 Largest Economic Countries in Southeast Asia.pdfMarket Analysis in the 5 Largest Economic Countries in Southeast Asia.pdf
Market Analysis in the 5 Largest Economic Countries in Southeast Asia.pdfRachmat Ramadhan H
ย 
๊งโค Aerocity Call Girls Service Aerocity Delhi โค๊ง‚ 9999965857 โ˜Ž๏ธ Hard And Sexy ...
๊งโค Aerocity Call Girls Service Aerocity Delhi โค๊ง‚ 9999965857 โ˜Ž๏ธ Hard And Sexy ...๊งโค Aerocity Call Girls Service Aerocity Delhi โค๊ง‚ 9999965857 โ˜Ž๏ธ Hard And Sexy ...
๊งโค Aerocity Call Girls Service Aerocity Delhi โค๊ง‚ 9999965857 โ˜Ž๏ธ Hard And Sexy ...Call Girls In Delhi Whatsup 9873940964 Enjoy Unlimited Pleasure
ย 

Kรผrzlich hochgeladen (20)

VidaXL dropshipping via API with DroFx.pptx
VidaXL dropshipping via API with DroFx.pptxVidaXL dropshipping via API with DroFx.pptx
VidaXL dropshipping via API with DroFx.pptx
ย 
Beautiful Sapna Vip Call Girls Hauz Khas 9711199012 Call /Whatsapps
Beautiful Sapna Vip  Call Girls Hauz Khas 9711199012 Call /WhatsappsBeautiful Sapna Vip  Call Girls Hauz Khas 9711199012 Call /Whatsapps
Beautiful Sapna Vip Call Girls Hauz Khas 9711199012 Call /Whatsapps
ย 
Customer Service Analytics - Make Sense of All Your Data.pptx
Customer Service Analytics - Make Sense of All Your Data.pptxCustomer Service Analytics - Make Sense of All Your Data.pptx
Customer Service Analytics - Make Sense of All Your Data.pptx
ย 
VIP High Class Call Girls Jamshedpur Anushka 8250192130 Independent Escort Se...
VIP High Class Call Girls Jamshedpur Anushka 8250192130 Independent Escort Se...VIP High Class Call Girls Jamshedpur Anushka 8250192130 Independent Escort Se...
VIP High Class Call Girls Jamshedpur Anushka 8250192130 Independent Escort Se...
ย 
VIP Call Girls Service Miyapur Hyderabad Call +91-8250192130
VIP Call Girls Service Miyapur Hyderabad Call +91-8250192130VIP Call Girls Service Miyapur Hyderabad Call +91-8250192130
VIP Call Girls Service Miyapur Hyderabad Call +91-8250192130
ย 
VIP Call Girls Service Charbagh { Lucknow Call Girls Service 9548273370 } Boo...
VIP Call Girls Service Charbagh { Lucknow Call Girls Service 9548273370 } Boo...VIP Call Girls Service Charbagh { Lucknow Call Girls Service 9548273370 } Boo...
VIP Call Girls Service Charbagh { Lucknow Call Girls Service 9548273370 } Boo...
ย 
Schema on read is obsolete. Welcome metaprogramming..pdf
Schema on read is obsolete. Welcome metaprogramming..pdfSchema on read is obsolete. Welcome metaprogramming..pdf
Schema on read is obsolete. Welcome metaprogramming..pdf
ย 
CebaBaby dropshipping via API with DroFX.pptx
CebaBaby dropshipping via API with DroFX.pptxCebaBaby dropshipping via API with DroFX.pptx
CebaBaby dropshipping via API with DroFX.pptx
ย 
April 2024 - Crypto Market Report's Analysis
April 2024 - Crypto Market Report's AnalysisApril 2024 - Crypto Market Report's Analysis
April 2024 - Crypto Market Report's Analysis
ย 
Delhi Call Girls Punjabi Bagh 9711199171 โ˜Žโœ”๐Ÿ‘Œโœ” Whatsapp Hard And Sexy Vip Call
Delhi Call Girls Punjabi Bagh 9711199171 โ˜Žโœ”๐Ÿ‘Œโœ” Whatsapp Hard And Sexy Vip CallDelhi Call Girls Punjabi Bagh 9711199171 โ˜Žโœ”๐Ÿ‘Œโœ” Whatsapp Hard And Sexy Vip Call
Delhi Call Girls Punjabi Bagh 9711199171 โ˜Žโœ”๐Ÿ‘Œโœ” Whatsapp Hard And Sexy Vip Call
ย 
100-Concepts-of-AI by Anupama Kate .pptx
100-Concepts-of-AI by Anupama Kate .pptx100-Concepts-of-AI by Anupama Kate .pptx
100-Concepts-of-AI by Anupama Kate .pptx
ย 
Invezz.com - Grow your wealth with trading signals
Invezz.com - Grow your wealth with trading signalsInvezz.com - Grow your wealth with trading signals
Invezz.com - Grow your wealth with trading signals
ย 
Log Analysis using OSSEC sasoasasasas.pptx
Log Analysis using OSSEC sasoasasasas.pptxLog Analysis using OSSEC sasoasasasas.pptx
Log Analysis using OSSEC sasoasasasas.pptx
ย 
Smarteg dropshipping via API with DroFx.pptx
Smarteg dropshipping via API with DroFx.pptxSmarteg dropshipping via API with DroFx.pptx
Smarteg dropshipping via API with DroFx.pptx
ย 
Midocean dropshipping via API with DroFx
Midocean dropshipping via API with DroFxMidocean dropshipping via API with DroFx
Midocean dropshipping via API with DroFx
ย 
Delhi Call Girls CP 9711199171 โ˜Žโœ”๐Ÿ‘Œโœ” Whatsapp Hard And Sexy Vip Call
Delhi Call Girls CP 9711199171 โ˜Žโœ”๐Ÿ‘Œโœ” Whatsapp Hard And Sexy Vip CallDelhi Call Girls CP 9711199171 โ˜Žโœ”๐Ÿ‘Œโœ” Whatsapp Hard And Sexy Vip Call
Delhi Call Girls CP 9711199171 โ˜Žโœ”๐Ÿ‘Œโœ” Whatsapp Hard And Sexy Vip Call
ย 
Unveiling Insights: The Role of a Data Analyst
Unveiling Insights: The Role of a Data AnalystUnveiling Insights: The Role of a Data Analyst
Unveiling Insights: The Role of a Data Analyst
ย 
VIP High Profile Call Girls Amravati Aarushi 8250192130 Independent Escort Se...
VIP High Profile Call Girls Amravati Aarushi 8250192130 Independent Escort Se...VIP High Profile Call Girls Amravati Aarushi 8250192130 Independent Escort Se...
VIP High Profile Call Girls Amravati Aarushi 8250192130 Independent Escort Se...
ย 
Market Analysis in the 5 Largest Economic Countries in Southeast Asia.pdf
Market Analysis in the 5 Largest Economic Countries in Southeast Asia.pdfMarket Analysis in the 5 Largest Economic Countries in Southeast Asia.pdf
Market Analysis in the 5 Largest Economic Countries in Southeast Asia.pdf
ย 
๊งโค Aerocity Call Girls Service Aerocity Delhi โค๊ง‚ 9999965857 โ˜Ž๏ธ Hard And Sexy ...
๊งโค Aerocity Call Girls Service Aerocity Delhi โค๊ง‚ 9999965857 โ˜Ž๏ธ Hard And Sexy ...๊งโค Aerocity Call Girls Service Aerocity Delhi โค๊ง‚ 9999965857 โ˜Ž๏ธ Hard And Sexy ...
๊งโค Aerocity Call Girls Service Aerocity Delhi โค๊ง‚ 9999965857 โ˜Ž๏ธ Hard And Sexy ...
ย 

Explaining the idea behind automatic relevance determination and bayesian interpolation

  • 1. Dr. Florian Wilhelm March 13th 2016 PyData Amsterdam P 1
  • 2. 2 Whatโ€˜s the best model to describe our data? And what does โ€žbestโ€œ actually mean?
  • 3. 3
  • 4. 4
  • 5. 5 Simple model โ€žGeneralityโ€œ Complex model โ€žBest Fitโ€œ Occamโ€˜s Razor: โ€žIt is vain to do with more what can be done with fewerโ€œ
  • 6. 6 Simple Model โ„‹1 Complex model โ„‹2 Space of all possible datasets ๐ท
  • 7. 7 Simple Model โ„‹1 Complex model โ„‹2 โ„‹1 fits only a small subset of ๐ท well
  • 8. 8 Simple Model โ„‹1 Complex model โ„‹2 โ„‹2 can fit large parts of ๐ท well
  • 9. 9 Prefer the model with high evidence for a given dataset Source: D. J. C. MacKay. Bayesian Interpolation. 1992
  • 10. 1. Model fitting: Assume โ„‹๐‘– is the right model and fit its parameters ๐’˜ with Bayes: ๐‘ƒ ๐’˜ ๐ท, โ„‹๐‘– = ๐‘ƒ ๐ท ๐’˜, โ„‹๐‘– ๐‘ƒ(๐’˜|โ„‹๐‘–) ๐‘ƒ(๐ท|โ„‹๐‘–) โ€œBusiness as usualโ€ 2. Model comparison: Compare different models with the help of their evidence ๐‘ƒ ๐ท โ„‹๐‘– and model prior ๐‘ƒ โ„‹๐‘– : ๐‘ƒ โ„‹๐‘– ๐ท โˆ ๐‘ƒ ๐ท โ„‹๐‘– ๐‘ƒ โ„‹๐‘– โ€œOccamโ€˜s razor at workโ€œ 10
  • 11. Marginalize & approximate: ๐‘ƒ ๐ท โ„‹๐‘– = ๐‘ƒ ๐ท ๐’˜, โ„‹๐‘– ๐‘ƒ ๐’˜ โ„‹๐‘– ๐‘‘๐’˜ ๐‘ƒ ๐ท โ„‹๐‘– โ‰… ๐‘ƒ ๐ท ๐’˜ ๐‘€๐‘ƒ, โ„‹๐‘– ๐‘ƒ ๐’˜ ๐‘€๐‘ƒ โ„‹๐‘– โˆ†๐’˜ ๐‘’๐‘ฃ๐‘–๐‘‘๐‘’๐‘›๐‘๐‘’ โ‰… ๐‘๐‘’๐‘ ๐‘ก ๐‘“๐‘–๐‘ก ๐‘™๐‘–๐‘˜๐‘’๐‘™๐‘–โ„Ž๐‘œ๐‘œ๐‘‘ ร— ๐‘‚๐‘๐‘๐‘Ž๐‘š ๐‘“๐‘Ž๐‘๐‘ก๐‘œ๐‘Ÿ 11 Occam factor: โˆ†๐’˜ โˆ† ๐ŸŽ ๐’˜ Source: D. J. C. MacKay. Bayesian Interpolation. 1992
  • 12. 12
  • 13. Given: ๏‚ง Dataset ๐ท = ๐‘ฅ ๐‘›, ๐‘ก ๐‘› with ๐‘› = 1 โ€ฆ ๐‘ ๏‚ง Set of (non-linear) functions ฮฆ = {๐œ™โ„Ž: ๐‘ฅ โŸผ ๐œ™(๐‘ฅ)} with โ„Ž = 1 โ€ฆ ๐‘€ Assumption: ๐‘ฆ ๐’™; ๐’˜ = โ„Ž=1 ๐‘€ ๐‘คโ„Ž ๐œ™โ„Ž(๐’™) , ๐‘ก ๐‘› = ๐‘ฆ ๐’™; ๐’˜ + ๐œ ๐‘›, where ๐œ ๐‘› is an additive noise with ๐’ฉ 0, ๐›ผโˆ’1 Task: Find min ๐’˜ โ€–ฮฆ๐’˜ โˆ’ ๐’•โ€–2 (Ordinary Least Squares) 13
  • 14. 14 Problem: Having too many features leads to overfitting! Regularization Assumption: โ€žWeights are smallโ€œ ๐‘ ๐’˜; ๐œ† ~๐’ฉ(0, ๐œ†โˆ’1 ๐•€) Task: Given ๐›ผ, ๐œ† find min ๐’˜ ๐›ผ ฮฆ๐’˜ โˆ’ ๐’• 2 + ๐œ† ๐’˜ 2
  • 15. 15 Consider each ๐›ผ๐‘–, ๐œ†๐‘– defining a model โ„‹๐‘– ๐›ผ, ๐œ† . Yes! That means we can use our Bayesian Interpolation to find ๐’˜, ๐œถ, ๐€ with the highest evidence! This is the idea behind BayesianRidge as found in sklearn.linear_model
  • 16. Consider that each weight has an individual variance, so that ๐‘ ๐’˜ ๐€ ~๐’ฉ 0, ฮ›โˆ’1 , where ฮ› = diag(๐œ†1, โ€ฆ , ๐œ† ๐ป), ๐œ†โ„Ž โˆˆ โ„+. Now, our minimization problem is: min ๐’˜ ๐›ผ ฮฆ๐’˜ โˆ’ ๐’• 2 + ๐’˜ ๐‘กฮ›๐’˜ 16 Pruning: If precision ๐œ†โ„Ž of feature โ„Ž is high, its weight ๐‘คโ„Ž is very likely to be close to zero and is therefore pruned. This is called Sparse Bayesian Learning or Automatic Relevance Determination. Found as ARDRegression under sklearn.linear_model.
  • 17. Crossvalidation can be used for the estimation of hyperparmeters but suffers from the curse of dimensionality (inappropriate for low-statistics). 17 Source: Peter Ellerton, http://pactiss.org/2011/11/02/bayesian-inference-homo-bayesianis/
  • 18. โ€ข Random 100 ร— 100 design matrix ฮฆ with 100 samples and 100 features โ€ข Weights ๐‘ค๐‘–, ๐‘– โˆˆ ๐ผ = 1, โ€ฆ , 100 , random subset J โŠ‚ ๐ผ with ๐ฝ = 10, and ๐‘ค๐‘– = 0, ๐‘– โˆˆ ๐ผJ ๐’ฉ(๐‘ค๐‘–; 0, 1 4), ๐‘– โˆˆ ๐ฝ โ€ข Target ๐’• = ฮฆ๐’˜ + ๐‚ with random noise ๐œˆ๐‘– โˆผ ๐’ฉ(0, 1 50) Task: Reconstruct the weights, especially the 10 non-zero weights! Source: http://scikit-learn.org/stable/auto_examples/linear_model/plot_ard.html#example-linear-model-plot-ard-py 18
  • 19. 19
  • 20. 20
  • 21. 21
  • 22. 22
  • 23. We have to determine the parameters ๐‘ค, ๐œ†, ๐›ผ for ๐‘ƒ ๐’˜, ๐€, ๐›ผ ๐’• = ๐‘ƒ ๐’˜ ๐’•, ๐€, ๐›ผ ๐‘ƒ ๐€, ๐›ผ ๐’• 1) Model fitting: For the first factor, we have ๐‘ƒ ๐’˜ ๐’•, ๐€, ๐›ผ ~๐’ฉ(๐, ฮฃ) with ฮฃ = ฮ› + ๐›ผฮฆ ๐‘‡ ฮฆ โˆ’1 , ๐ = ๐›ผฮฃฮฆT ๐ญ. 23
  • 24. 2) Model comparison: For the second factor, we have ๐‘ƒ ๐€, ๐›ผ ๐’• โˆ ๐‘ƒ ๐’• ๐€, ๐›ผ ๐‘ƒ ๐€ ๐‘ƒ ๐›ผ , where ๐‘ƒ ๐€ and ๐‘ƒ(๐›ผ) are hyperpriors which we assume uniform. Using marginalization, we have ๐‘ƒ ๐’• ๐€, ๐›ผ = ๐‘ƒ ๐’• ๐’˜, ๐›ผ ๐‘ƒ ๐’˜ ๐€ ๐‘‘๐’˜, i.e. marginal likelihood or the โ€œevidence for the hyperparameterโ€œ. 24
  • 25. Differentiation of the log marginal likelihood with respect to ๐œ†๐‘– and ๐›ผ as well as setting these to zero, we get ๐œ†๐‘– = ๐›พ๐‘– ๐œ‡๐‘– 2 , ๐›ผ = ๐‘ โˆ’ ๐‘– ๐›พ๐‘– ๐’• โˆ’ ฮฆ๐ 2 , with ๐›พ๐‘– = 1 โˆ’ ๐œ†๐‘–ฮฃ๐‘–๐‘–. These formulae are used to find the maximum points ๐€ ๐‘€๐‘ƒ and ๐›ผ ๐‘€๐‘ƒ. 25
  • 26. 1. Starting values ๐›ผ = ๐œŽโˆ’2(๐’•), ๐€ = ๐Ÿ 2. Calculate ฮฃ = ฮ› + ๐›ผฮฆ ๐‘‡ฮฆ โˆ’1 and ๐’˜ = ๐ = ๐›ผฮฃฮฆT ๐ญ 3. Update ๐œ†๐‘– = ๐›พ ๐‘– ๐œ‡ ๐‘– 2 and ๐›ผ = ๐‘โˆ’ ๐‘– ๐›พ ๐‘– ๐’•โˆ’ฮฆ๐ 2 where ๐›พ๐‘– = 1 โˆ’ ๐œ†๐‘–ฮฃ๐‘–๐‘– 4. Prune ๐œ†๐‘– and ๐œ™๐‘– if ๐œ†๐‘– > ๐œ† ๐‘กโ„Ž๐‘Ÿ๐‘’๐‘ โ„Ž๐‘œ๐‘™๐‘‘ 5. If not converged go to 2. Sklearn implementation: The parameters ๐›ผ1, ๐›ผ2 as well as ๐œ†1, ๐œ†2 are the hyperprior parameters for ๐›ผ and ๐€ with ๐‘ƒ ๐›ผ โˆผ ฮ“ ๐›ผ1, ๐›ผ2 โˆ’1 , ๐‘ƒ ๐œ†๐‘– โˆผ ฮ“ ๐œ†1, ๐œ†2 โˆ’1 . ๐ธ ฮ“ ๐›ผ, ๐›ฝ = ๐›ผ ๐›ฝ and ๐‘‰ ฮ“ ๐›ผ, ๐›ฝ = ๐›ผ ๐›ฝ2. 26
  • 27. Given a some new data ๐‘ฅโˆ—, a prediction for ๐‘กโˆ— is made by ๐‘ƒ ๐‘กโˆ— ๐’•, ๐€ ๐‘€๐‘ƒ, ๐›ผ ๐‘€๐‘ƒ = ๐‘ƒ ๐‘กโˆ— ๐’˜, ๐›ผ ๐‘€๐‘ƒ ๐‘ƒ ๐’˜ ๐’•, ๐€ ๐‘€๐‘ƒ, ๐›ผ ๐‘€๐‘ƒ ๐‘‘๐’˜ = ๐’ฉ ๐ ๐‘‡ ๐œ™ ๐‘ฅโˆ— , ๐›ผ ๐‘€๐‘ƒ โˆ’1 + ๐œ™ ๐‘ฅโˆ— ๐‘กฮฃ๐œ™ ๐‘ฅโˆ— . This is a good approximation of the predictive distribution ๐‘ƒ ๐‘กโˆ— ๐’• = ๐‘ƒ ๐‘กโˆ— ๐’˜, ๐€, ๐›ผ ๐‘ƒ ๐’˜, ๐€, ๐›ผ ๐’• ๐‘‘๐’˜ ๐‘‘๐€ ๐‘‘ฮฑ . 27
  • 28. 1. D. J. C. MacKay. Bayesian Interpolation. 1992 (โ€ฆ to understand the overall idea) 2. M. E.Tipping. Sparse Bayesian learning and the RelevanceVector Machine. June, 2001 (โ€ฆ to understand the ARD algorithm) 3. T. Fletcher. RelevanceVector Machines Explained. October, 2010 (โ€ฆ to understand the ARD algorithm in detail) 4. D.Wipf. A NewView of Automatic Relevance Determination. 2008 (โ€ฆ not as good as the ones above) Graphs from slides 7 and 9 were taken from [1] and the awesome tutorials of Scikit-Learn were consulted many times. 28
  • 29. 29