Week 7 - Generative Models - AC
Uploaded by
Erjon BrucajWeek 7 - Generative Models - AC
Uploaded by
Erjon BrucajECE 57000 – Artificial Intelligence
Generative Models
Murat Kocaoglu
The Need for Generative Models
• Goal: Generate new data that resembles training data.
• Generated synthetic data can help train better classifiers.
• Many creative/art applications.
• Central in modern AI with chatbots, image and video generation.
What do we need to generate data?
Joint distribution of data p(x,y)
- sampling from p(x,y) is not always easy
- learning p(x,y) is not always easy
- representing p(x,y) is not always easy!
What does p(x,y) look like? For discrete variables:
X\Y 0 1 2 For 3-state X, Y, joint has 9 non-negative numbers.
0 0.1 0.2 0.2
For n-state X,Y joint has n2 non-negative numbers.
1 0.01 0.1 0.2
2 0.04 0.05 0.1
How about m variables each with n states?
X
sha1_base64="ZM4Jlrj7HUFE3KghPebLWG20fwI=">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</latexit>
<latexit
Z
representations.
This is a Bayesian network!
= p(x)p(z|x)p(y|z)
p(x, y, z) = p(x)p(z|x)p(y|x, z)
n + n2 + n2
# Parameters:
n + n2 + n3
Structured Probability Distributions
Using domain knowledge, we can sometimes simplify this joint
Suppose we know X, Y are independent given Z; n states each.
Probabilistic Graphical Models
Use directed or undirected graphs to represent structure in probability
distributions.
Z Z
X Y X Y
T T
Bayesian Network Markov Random Field
Vertices (nodes) represent random variables.
Missing edges encode conditional independences in the joint.
Probabilistic Graphical Models
Use directed or undirected graphs to represent structure in probability
distributions.
Z Z
X Y X Y
T T
Bayesian Network Markov Random Field
Vertices (nodes) represent random variables.
Missing edges encode conditional independences in the joint.
Bayesian Networks
A Bayesian network represents factorization of a joint distribution
encoded as a directed acyclic graph (DAG).
Cloud presence increases likelihood of rain
C
- C• R
R S Cloud presence decreases likelihood of
sprinkler being turned on
W - C• S
Sprinkler Example Sprinkler and rain both affect wetness of grass
- R• W, S• W
[Do you mean “causes”?]
R
C
Sprinkler Example
S
sha1_base64="Lzsa4MAFTYjzK+ozN0Oc0igLVBA=">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</latexit>
<latexit
sha1_base64="OEemnl7KiJClr8oh8CLhESGozVM=">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</latexit>
<latexit sha1_base64="floO28jalk6B28fiYhRghazsBrw=">AAAJt3icjVbNjts2EFbSv7X7k0177IWtvcBuoHXtJEhboAtksT3k0ALuIpsEsIwFRdE2YYpkSCq2w+gJ+jS9tk/St+lQlteSbAOVLXs0881HznA4VKw4M7bf//fe/Y8+/uTTz45a7c+/+PKrB8cPv35lZKYJvSGSS/0mxoZyJuiNZZbTN0pTnMacvo7nV97++h3Vhknx0q4UHad4KtiEEWxBdXt8ok5JqEMTLs7QBYKHM3WqP/hfU/wuPoDx7Pa40+/1iwvtCoNS6ATlNbx9ePQ+SiTJUios4diY0aCv7NhhbRnhNG9HmaEKkzme0hGIAqfUjF0RT45OQJOgidRwC4sKbdXD4dSYVRoDMsV2Zpo2rzxks7P0ztSOBF0QmaZYJC6yM/oC86nUzGNchDWOGXFbVd6uk20MjOT7DXWHUWYnP40dEyqzVJB1mJOMIyuRXxqUME2J5SsQMAECRhCZwSyIhQWsMSXvmDJlzpbrpMFQEIDUFEJcceoUx0zkRYSl3pX/uXtZCiMD40EZjGswpaWShnlDPiqVYzesaGtwTiF/FeBvxXMNQqSWnGO9qsCu7nSNiSd0wsSecSr6LcuvB8BQcVmqGuDLrbIxJghYz+sUpW7rfl2C6lXDREIVgnJRVKvou+LjxbwOIywfPRm7zgCVHp3HKOJ0YiPYnRZ1nkSaTWe213ATFT8hbcldGeYwT60kLZu/rxepyWKCN8mo6NOMW6blIt+7qdpNrd+gpsEAQCslNw300oBEGwEqRXLnutHQsO758CqP9qJwZmWKVe4uQfj9cniHqsNMETvWfvruUZGjiqaB9UkrodGd7H0eNZFg8veGMo3l0p2fn+cluIEmTPsOtwZ3t+iu3CkJD9zA5Ra5H7iegqwOv2+m1eEPMlrM+Jpvi9mJhFSzKRvZ3JtTcjCnOyF5a42/lil0sHy9xFmsoXH4QtNsGSpsoUEKswswM6yoCWNgmWqZiWQPhmAO7fvkgGtCoXcVR6UJi5ma0DuUci+lFocTZsNNa2Ri6sk8E/C59klLSSbsD72iz8AJ69brE6JE40WImBBUIwj0Ak7TpyQFMs5DIRMKp56AY8KaCwcHz4nvWBvSluf4/5S7bLtZqM2/mvmpxmrGyLJZHHPYq64793sWNdb2Oh8NxrClfR+AZLnrvDPoNv29MY7TAhpZurTOj+q7ibvJnbc4+ApwbR67sZRzi2MD52fxiiAYgXhBA8nQGeyTNrytDJrvJrvCq8e9wbNe/4+nnee/lO8tR8G3wffBaTAIfgyeBy+CYXATkODP4K/g7+Cf1s+t29akNVtD798rfb4Jalfr7X8e2o1m</latexit>
<latexit
Bayesian Networks
p(s|c, r) = p(s|c)
&
sha1_base64="WLLD2jybzKfHSohDiwnDqGQAt4o=">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</latexit>
<latexit
encoded as a directed acyclic graph (DAG).
Factorization of the joint distribution:
Compare with chain rule for any distribution:
p(w|r, s, c) = p(w|r, s)
p(c, r, s, w) = p(c)p(r|c)p(s|c)p(w|r, s)
A Bayesian network represents factorization of a joint distribution
p(c, r, s, w) = p(c)p(r|c)p(s|c, r)p(w|r, s, c)
R
C
Sprinkler Example
S
sha1_base64="Lzsa4MAFTYjzK+ozN0Oc0igLVBA=">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</latexit>
<latexit
sha1_base64="SpjxmEBHMat/YOKEz5YihENbuLQ=">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</latexit>
<latexit
sha1_base64="OEemnl7KiJClr8oh8CLhESGozVM=">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</latexit>
<latexit sha1_base64="floO28jalk6B28fiYhRghazsBrw=">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</latexit>
<latexit
Bayesian Networks
S ?? R |C
p(s|c, r) = p(s|c)
&
sha1_base64="WLLD2jybzKfHSohDiwnDqGQAt4o=">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</latexit>
<latexit
encoded as a directed acyclic graph (DAG).
sha1_base64="dfBnSHbEyRCkYsbD0uDYiv9cFto=">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</latexit>
<latexit
Factorization of the joint distribution:
W ?? C |R, S
Compare with chain rule for any distribution:
p(w|r, s, c) = p(w|r, s)
p(c, r, s, w) = p(c)p(r|c)p(s|c)p(w|r, s)
A Bayesian network represents factorization of a joint distribution
p(c, r, s, w) = p(c)p(r|c)p(s|c, r)p(w|r, s, c)
R
C
Sprinkler Example
S
sha1_base64="OEemnl7KiJClr8oh8CLhESGozVM=">AAAJu3icjVbNjts2EFbSv7X7t2mPubC1F8gGWtdOgraHLJDF9pBDC7iLbBLAMhYURdusKZIgqdgOV8/Qp+k1eY6+TYeyvJZkG6j8N5755iNnNJxRrDgztt//9979Tz797PMvjlrtL7/6+ptvjx9899rITBN6TSSX+m2MDeVM0GvLLKdvlaY4jTl9E88vvf3NO6oNk+KVXSk6TvFUsAkj2ILq5vhUPSKhDk24OEXnCP6AQt/6b3MLBvhd3HozOb057vR7/eJCu8KgFDpBeQ1vHhy9jxJJspQKSzg2ZjToKzt2WFtGOM3bUWaowmSOp3QEosApNWNXxJSjE9AkaCI1fIRFhbbq4XBqzCqNAZliOzNNm1cestlZemdqR4IuiExTLBIX2Rl9iflUauYxLsIax4y4rSpv18k2Bkby/Ya6wyizk1/HjgmVWSrIOsxJxpGVyN8elDBNieUrEDABAkYQmcEuiIWbWGNK3jFlypwt10mDpSAAqSmEuOLUKY6ZyIsIS70rf3P3qhRGBtaDUhjXYEpLJQ3zhnxUKsduWNHW4JxC/irA34v/NQiRWnKO9aoCu7zTNTae0AkTe9ap6Lcsvx0AQ8VlqWqAL7bKxpogYD2vU5S6rftVCapXDRMJVQjKRVGtoh+KlxfzOoywfPR07DoDVHp0nqCI04mN4IRa1HkaaTad2V7DTVT8hLQld2WZwzy1krRs/r5epCaLCd4ko6JPM26Zlot876FqN7X+gJoGAwCtlNw00EsDEm0EqBTJnetGQ8O6Z8PLPNqLwpmVKVa5uwDhj4vhHaoOM0XsWPvtu8dFjiqaBtYnrYRGd7L3edxEgsl/NpRpLJfu7OwsL8ENNGHad7g1uLtFd+VOSXjgBi63yP3A9RZkdfl9O60uf5DRYsbXfFvMTiSkmk3ZyObenJKDOd0JyVtr/LVMoYPl6yXOYg2NwxeaZstQYQsNUphdgJlhRU0YA8tUy0wkezAEc2jfJwdcEwq9qxiXJix2CtMQHEq5l1KLwwmz4aY1MjH1ZJ4J+Fz7pKUkE/anXtFnYMq69f0JUaLxIkRMCKoRBHoO0/QZSYGM81DIhMLUEzAmrDl3MHhOfMfakLY8x/+n3GXbzUJt/9XMTzVWM0aWzeKYw1l13bk/s6hxb6/y0WAMR9r3AUiWu8o7g27T3xvjOC2gkaVL6/yqvpu469x5i4O3ANfm2I2lnFscG5ifxSOCYATiBQ0kQ2dwTtrwtDJoPpvsCq+f9AY/9/p/Puu8eF4+txwFD4Mfg0fBIPgleBG8DIbBdUCCv4N/gg/Bx9Z5i7T+avE19P690uf7oHa1sv8AzQOOuw==</latexit>
<latexit sha1_base64="floO28jalk6B28fiYhRghazsBrw=">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</latexit>
<latexit
Bayesian Networks
# of param:
# of param:
n
n
encoded as a directed acyclic graph (DAG).
+
+
n2
n2
+ n3
+ n2
+
+
n4
n3
p(c, r, s, w) = p(c)p(r|c)p(s|c)p(w|r, s)
=
=
A Bayesian network represents factorization of a joint distribution
p(c, r, s, w) = p(c)p(r|c)p(s|c, r)p(w|r, s, c)
sha1_base64="urN87WcS4qc7JO2JBTQ33AS1JiU=">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</latexit>
<latexit
O(n3 )
M D
Collider Bias C
Conditioning on common children can make otherwise independent
variables dependent!
Scenario:
Having observed that the child has blue eyes, further observing that
mom has brown eyes indicate that the dad must have blue eyes.
- because blue is recessive and brown is dominant gene.
Example:
X, Y independent and
Given Z=5, observing X=1 sets Y=4 vs. observing X=2 sets Y=3.
Bayesian Networks
Can we check if two variables are conditionally independent
given a third, just by looking at the graph?
YES!
In polynomial time?
YES!! (Bayes Ball Algorithm)
Global Markov Property.
X and Y are independent given Z if X and Y are d-
separated given Z in the graph.
Bayesian Networks
d-separation:
We say X, Y are d-separated given Z on a DAG D if there does
not exist any d-connecting path between X, Y given Z.
d-connecting path:
We say a path q between X and Y is d-connecting given Z if
- no non-collider is in Z, and
- all colliders are either in Z or in An(Z)
collider:
A triple <X,Y,Z> is a collider if X•Y and Y•Z.
U
X
Y
d-separation
Bayesian Network
Z
V
sha1_base64="i/X9iTI69IipBy8Og/Bm3avf3wg=">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</latexit>
<latexit
U ?? V |;
d-separation
14
U
U
X
X
Y
Y
d-separation
Bayesian Network
Z
Z
V
V
sha1_base64="AM1jmnxMt+8c925+W0QJpT1IXSU=">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</latexit>
<latexit sha1_base64="i/X9iTI69IipBy8Og/Bm3avf3wg=">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</latexit>
<latexit
U 6?? V |Y
U ?? V |;
d-separation
15
U
U
U
X
X
X
Y
Y
Y
d-separation
Bayesian Network
Z
Z
Z
V
V
V
sha1_base64="QCtj0UdDhDFXGVWmX6zBsh88VHc=">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</latexit>
<latexit sha1_base64="AM1jmnxMt+8c925+W0QJpT1IXSU=">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</latexit>
<latexit sha1_base64="i/X9iTI69IipBy8Og/Bm3avf3wg=">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</latexit>
<latexit
U 6?? V |Y
U ?? V |;
d-separation
U ?? V |Y, Z
16
U
U
U
U
X
X
X
X
T
Y
Y
Y
Y
d-separation
Bayesian Network
Z
Z
Z
Z
V
V
V
V
sha1_base64="qUQDMlJdl/8YphcaxdALCdyKU/U=">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</latexit>
<latexit sha1_base64="QCtj0UdDhDFXGVWmX6zBsh88VHc=">AAADhHicdVJNbxMxEHWzUMry0RaOXAxbJA4l2m1o4YRS9cIFKUgkLcSheB0nseK1LdtLCZb/B1f4V/wbvOkKJdswlqWnN2/G9vPkijNj0/TPViu6dXv7zs7d+N79Bw939/YfDYwsNaF9IrnUFzk2lDNB+5ZZTi+UprjIOT3P52dV/vwb1YZJ8dEuFB0VeCrYhBFsA/UFEeb63g28+3T42V/uJWk7XQa8CbIaJKCO3uV+6ysaS1IWVFjCsTHDLFV25LC2jHDqY1QaqjCZ4ykdBihwQc3ILa/t4fPAjOFE6rCFhUt2tcLhwphFkQdlge3MNHMVuSk3LO3kzcgxoUpLBbk+aFJyaCWsPIBjpimxfBEAJpqFu0IywxoTG5yK4xgJekVkUWAxdoiJMVXQO6SoVujpclXQr8sI88POyCUZrCuSI4g4nVgU7Lcw6SDNpjPbbpSJlTohbd175Zj/91mzw7L5jzW7nSlzglX1y+t8UXLLtLzyG+2Mm2z1NabRIQitlNw01N9NQLTxQKWId+4A9Qw7eNk782ijCpdWFlh5dxrA+9PeP1UcZjJrTuBNMDhqZyft9MOrpNutp3MHPAHPwAuQgdegC96BHugDAjT4CX6B39F2dBh1ouNraWurrnkM1iJ6+xf2NCV5</latexit>
<latexit sha1_base64="AM1jmnxMt+8c925+W0QJpT1IXSU=">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</latexit>
<latexit sha1_base64="i/X9iTI69IipBy8Og/Bm3avf3wg=">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</latexit>
<latexit
U 6?? V |Y
U 6?? V |T
U ?? V |;
d-separation
U ?? V |Y, Z
17
sha1_base64="tbpx/J/eW6v43u3GHWcCxU+bM+M=">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</latexit>
<latexit
Factorization Property.
p(x1 , x2 . . . xn ) =
n
Y
i=1
p(xi |pa(xi ))
Markov Properties of Bayesian Networks
R
C
W
sha1_base64="tbpx/J/eW6v43u3GHWcCxU+bM+M=">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</latexit>
<latexit
S
Factorization Property.
sha1_base64="floO28jalk6B28fiYhRghazsBrw=">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</latexit>
<latexit
p(x1 , x2 . . . xn ) =
n
Y
i=1
p(xi |pa(xi ))
Markov Properties of Bayesian Networks
p(c, r, s, w) = p(c)p(r|c)p(s|c)p(w|r, s)
sha1_base64="KSamj5ybh7PIr8ER3u327/ybZv8=">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</latexit>
<latexit
Local Markov Property.
X ?? N onDesc(X) |P aX
Markov Properties of Bayesian Networks
X is independent of its non-descendants (except parents) given its parents.
R
C
W
sha1_base64="KSamj5ybh7PIr8ER3u327/ybZv8=">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</latexit>
<latexit
S
Local Markov Property.
sha1_base64="neLTJwRsbBOX5xh7xRnUEXHOm3o=">AAAJnHicjVZtjxo3EN6kbwd9S9qPlSq3cFIV7VFIorYfelJOlw+R2kj0mrtEYlHk9Rqw8NqW7c1BLP+Nfm3/Vv9Nx8ty7C4gdWHFMPPMY894PHaqODN2OPz33v0PPvzo409OOt1PP/v8iy8fPPzqxshCE3pNJJf6TYoN5UzQa8ssp2+UpjhPOX2dLi+D/fU7qg2T4pVdKzrN8VywGSPYgipJCHNX3v3p3aV/+6A3HAzLB+0Lo0roRdUzfvvw5H2SSVLkVFjCsTGT0VDZqcPaMsKp7yaFoQqTJZ7TCYgC59RMXTlpj05Bk6GZ1PAKi0pt3cPh3Jh1ngIyx3Zh2ragPGazi/zO1E0EvSUyz7HIXGIX9AXmc6lZwLgEa5wy4nYq322SbQ2M+MOGpsOksLNfpo4JVVgqyCbMWcGRlSjkH2VMU2L5GgRMgIARRBYwC2JhlRpM2TumTJWz1SZpMBQEIDWFENecOsUxE76MsNK76te7V5UwMTAerPW0AVNaKmlYMPhJpZy6cU3bgHMK+asBfy//NyBEask51usa7PJO15p4RmdMHBinpt+xPD8ChoorctUCX+yUrTFBwHrZpKh0O/erCtSsGiYyqhCUi6JaJd+VnyD6JowwP3kydb0Rqjx6j1HC6cwmsAUt6j1JNJsv7KDlJmp+QtqKuzbMcZ5GSVq2fN8sUlOkBG+TUdPnBbdMy1t/cFN129qwQU2LAYBWSm5a6JUBibYCVIp45/rJ2LD+2fjSJwdRuLAyx8q7CxBeXozvUE2YKWPHOkzfPSpzVNO0sCFpFTS5k4PPozYSTOHdUuapXLmzszNfgVtownTocBtwf4fuy72SCMAtXO6Qh4GbKcj68IdmWh/+KKPFjG/4dpi9SEg9m7KVzYM5JUdzuhdSsDb4G5lCR8s3SJylGhpHKDTNVrHCFhqkMPsAs8CKmjgFlrmWhcgOYAjm0L5Pj7hmFHpXeR6auJypiYNDJQ9yanE8YzbetkYm5oEsMAGf6552lGTC/jgo+ww6R26zPjHKNL6NEROCagSBnsNp+pTkQMZ5LGRG4dQTcExYc+7g4DkNHWtL2gkc/59yn20/C4351zM/11gtGFm1i2MJe9X1l2HPotbaXvnJaApbOvQBSBZcInqjfts/GNM0L6GJpSvrwqihm7hr74LFwVeAa/vYTaVcWpwaOD/LK4JgBOIFDSRDF7BPunBbGbXvJvvCzePB6KfB8I+nvWe/VveWk+ib6Pvoh2gU/Rw9i15E4+g6IpGK/or+jv7pfNt53vmt83IDvX+v8vk6ajydm/8AMMOFpA==</latexit>
<latexit
X ?? N onDesc(X) |P aX
R ?? S |C
Markov Properties of Bayesian Networks
X is independent of its non-descendants (except parents) given its parents.
R
C
W
sha1_base64="KSamj5ybh7PIr8ER3u327/ybZv8=">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</latexit>
<latexit
S
Local Markov Property.
sha1_base64="neLTJwRsbBOX5xh7xRnUEXHOm3o=">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</latexit>
<latexit
X ?? N onDesc(X) |P aX
R ?? S |C
Note that further conditioning on W would make them dependent!
Markov Properties of Bayesian Networks
X is independent of its non-descendants (except parents) given its parents.
A
R
C
sha1_base64="F8QMtpX6iXOVlcEcXgE53YmUEjg=">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</latexit>
<latexit
S
D
W
Global Markov Property.
B
X ?? Y |Z () (X ?? Y |Z )D
Markov Properties of Bayesian Networks
X and Y are independent given Z if X and Y are d-separated given Z in the graph.
A
R
C
sha1_base64="F8QMtpX6iXOVlcEcXgE53YmUEjg=">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</latexit>
<latexit
S
D
W
Global Markov Property.
sha1_base64="lWaFC/a3uRKxrbfAOe+huVNOp1s=">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</latexit>
<latexit
X ?? Y |Z () (X ?? Y |Z )D
A 6?? S |;
?
sha1_base64="m2HwyKMflPotP1HxIgb/JNC8wtI=">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</latexit>
<latexit
Markov Properties of Bayesian Networks
A ?? S |;
X and Y are independent given Z if X and Y are d-separated given Z in the graph.
A
R
C
sha1_base64="F8QMtpX6iXOVlcEcXgE53YmUEjg=">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</latexit>
<latexit
S
D
W
Global Markov Property.
sha1_base64="lWaFC/a3uRKxrbfAOe+huVNOp1s=">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</latexit>
<latexit
X ?? Y |Z () (X ?? Y |Z )D
A 6?? S |;
Markov Properties of Bayesian Networks
X and Y are independent given Z if X and Y are d-separated given Z in the graph.
A
R
C
sha1_base64="F8QMtpX6iXOVlcEcXgE53YmUEjg=">AAAJtHicjVbNjts2EFbSv7X7k0177IWNvUAaeF07CdIeukAWm0MOLeAussm2lrGgKNomTJEESWXtELz3aXptX6Vv06EsryXZBipb8Hjmm4+c4XDIRHFm7GDw7737H338yaefHbXan3/x5VcPjh9+/dbIXBN6RSSX+jrBhnIm6JVlltNrpSnOEk7fJYuLYH/3nmrDpHhjV4pOMjwTbMoItqC6OX4UE+auvfvduz88itl0ih5XVd/fvLo57gz6g+JBu8KwFDpR+YxuHh59iFNJ8owKSzg2ZjwcKDtxWFtGOPXtODdUYbLAMzoGUeCMmokrgvHoBDQpmkoNr7Co0FY9HM6MWWUJIDNs56ZpC8pDNjvP7kztWNBbIrMMi9TFdk5fYz6TmgWMi7HGCSNuq/LtOtnGwIjfb6g7jHM7/WnimFC5pYKsw5zmHFmJwrqglGlKLF+BgAkQMILIHGZBLKxejSl9z5Qpc7ZcJw2GggCkphDiilOnOGbCFxGWelf+evemFMYGxoMamNRgSkslDQsGPy6VEzeqaGtwTiF/FeAvxf8ahEgtOcd6VYFd3OkaE0/plIk941T0W5ZXB8BQcXmmGuDzrbIxJghYL+oUpW7rflmC6lXDREoVgnJRVKv4u+ITRF+HEebHzyauM0SlR+cpijmd2hi2pkWdZ7Fms7ntN9xExU9IW3JXhjnMUytJyxYf6kVq8oTgTTIq+iznlml56/duqnZTGzaoaTAA0ErJTQO9NCDRRoBKEe9cNx4Z1j0dXfh4LwrnVmZYeXcOwq/noztUHWaK2LEO03dPihxVNA1sSFoJje/k4POkiQRTeDeUWSKX7vT01JfgBpowHTrcGtzdortypyQCcAOXW+R+4HoKsjr8vplWhz/IaDHja74tZicSUs2mbGRzb07JwZzuhBSsNf5aptDB8g0SZ4mGxhEKTbNlT2ELDVKYXYCZY0VNLwGWmZa5SPdgCObQvk8OuKYUeldxTppeMVPTCw6l3M+oxb0ps71Na2RiFsgCE/C59klLSSbsD/2iz6Az5Nbr00Opxrc9xISgGkGgZ3CaPicZkHHeEzKlcOoJOCasOXNw8JyEjrUhbQWO/0+5y7abhdr8q5mfaazmjCybxbGAveq6i7BnUWNtL/14OIEtHfoAJMtd+s6w2/QPxiTJCmhs6dK6MGroJu7Ku2Bx8BXg2jx2EykXFicGzs/iiiAYgXhBA8nQOeyTNtxWhs27ya7w9ml/+KI/+O155+XP5b3lKPo2ehQ9jobRj9HL6HU0iq4iEv0Z/RX9Hf3TetGKW6RF19D790qfb6La0xL/AbgJjrQ=</latexit>
<latexit
S
D
W
Global Markov Property.
sha1_base64="UyXP/Giw7DspauG76MeQgHYgLdQ=">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</latexit>
<latexit
X ?? Y |Z () (X ?? Y |Z )D
sha1_base64="PI25CBT40nzcDjmh3oxAYJS7QNo=">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</latexit>
<latexit
Markov Properties of Bayesian Networks
A 6?? S |R ? A ?? S |R
X and Y are independent given Z if X and Y are d-separated given Z in the graph.
A
R
C
sha1_base64="F8QMtpX6iXOVlcEcXgE53YmUEjg=">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</latexit>
<latexit
S
D
W
Global Markov Property.
sha1_base64="UyXP/Giw7DspauG76MeQgHYgLdQ=">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</latexit>
<latexit
X ?? Y |Z () (X ?? Y |Z )D
A 6?? S |R
Markov Properties of Bayesian Networks
X and Y are independent given Z if X and Y are d-separated given Z in the graph.
A
R
C
sha1_base64="F8QMtpX6iXOVlcEcXgE53YmUEjg=">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</latexit>
<latexit
S
D
W
Global Markov Property.
sha1_base64="CTHg2KwGiB5TZbb086rSy96cUyw=">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</latexit>
<latexit
X ?? Y |Z () (X ?? Y |Z )D
A ?? S |R, C
Markov Properties of Bayesian Networks
X and Y are independent given Z if X and Y are d-separated given Z in the graph.
A
R
C
sha1_base64="F8QMtpX6iXOVlcEcXgE53YmUEjg=">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</latexit>
<latexit
S
D
W
Global Markov Property.
sha1_base64="JUnrLXaind6EzMrP1Yfr1osrGVA=">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</latexit>
<latexit
X ?? Y |Z () (X ?? Y |Z )D
A ?? B |;
Markov Properties of Bayesian Networks
X and Y are independent given Z if X and Y are d-separated given Z in the graph.
AR
C
sha1_base64="F8QMtpX6iXOVlcEcXgE53YmUEjg=">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</latexit>
<latexit
S
D
W
Global Markov Property.
sha1_base64="l4kDSAh+t13ILfhmlYC8uYHwof0=">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</latexit>
<latexit
X ?? Y |Z () (X ?? Y |Z )D
A 6?? B |D
Markov Properties of Bayesian Networks
X and Y are independent given Z if X and Y are d-separated given Z in the graph.
AR
C
sha1_base64="F8QMtpX6iXOVlcEcXgE53YmUEjg=">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</latexit>
<latexit
S
D
W
Global Markov Property.
sha1_base64="l4kDSAh+t13ILfhmlYC8uYHwof0=">AAAJnXicjVZfjxo3EN8k/XPQf0n7mIe4hZOqiKOQRGkfclKul0p5SCt6yuUisSjyeg1YeG3X9uYglj9HXtuP1W/T8bIcuwtIXVgxzPzmZ894PHaiODN2MPj31u07n3z62edHrfYXX3719Td37337xshcE3pJJJf6bYIN5UzQS8ssp2+VpjhLOL1KFufBfvWeasOkeG1Xik4yPBNsygi2oJrEgjB35t2v3r3w7+52Bv1B8aBdYVgKnah8Ru/uHX2IU0nyjApLODZmPBwoO3FYW0Y49e04N1RhssAzOgZR4IyaiStm7dExaFI0lRpeYVGhrXo4nBmzyhJAZtjOTdMWlIdsdp7dmNqxoNdEZhkWqYvtnL7EfCY1CxgXY40TRtxW5dt1so2BEb/fUHcY53b6y8QxoXJLBVmHOc05shKFBUAp05RYvgIBEyBgBJE5zIJYWKYaU/qeKVPmbLlOGgwFAUhNIcQVp05xzIQvIiz1rvz17nUpjA2MFxa7BlNaKmlYMPhxqZy4UUVbg3MK+asAXxX/axAiteQc61UFdn6ja0w8pVMm9oxT0W9ZXhwAQ8XlmWqAz7bKxpggYL2oU5S6rftFCapXDRMpVQjKRVGt4u+LTxB9HUaYHz+euM4QlR6dRyjmdGpj2IMWdR7Hms3mtt9wExU/IW3JXRnmME+tJC1bfKgXqckTgjfJqOiznFum5bXfu6naTW3YoKbBAEArJTcN9NKARBsBKkW8c914ZFj3ZHTu470onFuZYeXdGQi/n41uUHWYKWLHOkzfPSxyVNE0sCFpJTS+kYPPwyYSTOHdUGaJXLqTkxNfghtownTocGtwd4vuyp2SCMANXG6R+4HrKcjq8PtmWh3+IKPFjK/5tpidSEg1m7KRzb05JQdzuhNSsNb4a5lCB8s3SJwlGhpHKDTNlj2FLTRIYXYBZo4VNb0EWGZa5iLdgyGYQ/s+PuCaUuhdxYFoesVMTS84lHI/oxb3psz2Nq2RiVkgC0zA59rHLSWZsD/1iz6DTpFbr08PpRpf9xATgmoEgZ7CafqEZEDGeU/IlMKpJ+CYsObUwcFzHDrWhrQVOP4/5S7bbhZq869mfqaxmjOybBbHAvaq6y7CnkWNtb3w4+EEtnToA5Asd+E7w27TPxiTJCugsaVL68KooZu4S++CxcFXgGvz2E2kXFicGDg/iyuCYATiBQ0kQ+ewT9pwWxk27ya7wptH/eHT/uDPJ53nz8p7y1F0P/oh+jEaRj9Hz6OX0Si6jEj0V/Qx+jv6p/Wg9VvrVeuPNfT2rdLnu6j2tK7+AyR4hfs=</latexit>
<latexit
X ?? Y |Z () (X ?? Y |Z )D
A 6?? B |D
- write down factorization
Markov Properties of Bayesian Networks
- show p(a,b|d)=/=p(a|d)p(b|d)
Imagine trying to infer this from joint:
- marginalize out all but A, B, D
X and Y are independent given Z if X and Y are d-separated given Z in the graph.
What about images?
How many numbers represent a joint distribution of a 256x256
image with 8-bit pixels?
What about images?
How many numbers represent a joint distribution of a 256x256
image with 8-bit pixels?
A 256x256 image has 65536 pixels each with 28=256 values.
What about images?
How many numbers represent a joint distribution of a 256x256
image with 8-bit pixels?
A 256x256 image has 65536 pixels each with 28=256 values.
The joint distribution would have 25665536 numbers!
- there are 1080 atoms in the universe
What about images?
How many numbers represent a joint distribution of a 256x256
image with 8-bit pixels?
A 256x256 image has 65536 pixels each with 28=256 values.
The joint distribution would have 25665536 numbers!
- there are 1080 atoms in the universe
Also, it is hard to imagine any persistent independence structure
between pixels.
Clearly, we need a better way.
Mixture of Gaussians
Treating image as a continuous variable, we can try writing it as
mixture of Gaussians. However, this is not suitable for image
data, will give blurry images.
https://medium.com/@tilak559/gaussian-mixture-modeling-gmm-1cb5f00a97d9
Noise
Gaussian
Noise N
sha1_base64="d6XUVd8yjlOwxLR0JfEw45mwO1k=">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</latexit>
sha1_base64="d6XUVd8yjlOwxLR0JfEw45mwO1k=">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</latexit><latexit
sha1_base64="d6XUVd8yjlOwxLR0JfEw45mwO1k=">AAAHBnicjVXdjhs1FJ4WGtrhr6WX3LhkVyooDZkt6vZmpdKCoBJIy7bbVsqEyuM5k5h47KntaTa1fM8Nt/AW3CFueQ0egnfgzE+7mUmQmGjkL+d857N9fHwmKQQ3djL5+8LFt96+NHjn8pXw3ffe/+DDq9c+emJUqRmcMiWUfpZQA4JLOLXcCnhWaKB5IuBpsnxQ+Z++BG24ko/tuoBZTueSZ5xRi6ZHe9ne86vDyXhSP2QbRC0YBu1z/PzapX/iVLEyB2mZoMZMo0lhZ45qy5kAH8algYKyJZ3DFKGkOZiZq9fqyT5aUpIpja+0pLZuRjiaG7POE2Tm1C5M31cZd/mmpc3uzhyXRWlBsmairBTEKlJtnKRcA7NijYAyzXGthC2opsxiejqzVNpWKWE6W3GWL191LaZMGC2qRDb2iiF4oqleYza0Wpltu1nQAswoQYm5VqVM/5uTAlO6PiczauRGjArW4nEOlo4ybkeFMrxicTlHrUoI5Vx4pVBc2s/Hxq4FkCPiGNd4PiOSaroaES4laIKbOcLD/oLlKCXESKoU8FAk5tCaI+d96MMwlrBiKs+pTF2M2fHTaFaDRZK5YYSkDcZnLn4B6SPv4kWlGIM0pYaK7GLzoqQavN/vSmpIG0kLZ1jDLq6rwtXmYTTr6rsYfDt54r7u++zZudLW0lycgqj9Li6qcqWC7OTYlerSfjzYQeQyhYLgYgrQRXyj/lWwR2PcT2/PcDGkjRgekFhAZmO8mJYMb8eazxd2THpxciNQKtuKb8zzf4VONs4r2ZGUouM/9rUmJr6Rm/XZO+k3X9M/7dEftoeFlesebmXwe6qX6qV3exucvVuNtd2GXYDC+nEpZFzWde7dV+e4wxGAwt59Vw8dD94kJQReLu8evIEdRqHV64vk3fHGn3C/R1NZTcChI1AVuV56d9KMfXHsyHkdV4OOtx29e9yCjhc7bJk3XcZ9eY5725M/YXfDi1bt7w0OO+3qrGnBaKxzCc0tK9QKa6lqFYkoMfzkm/ve3bkziiJ8Dw99iB+IqP852AZPDsYR4h8Ohvfut5+Ky8HHwSfBzSAKDoN7wbfBcXAasGAe/BL8Gvw2+Hnw++CPwZ8N9eKFNuZ60HkGf/0LWMeJog==</latexit><latexit
sha1_base64="d6XUVd8yjlOwxLR0JfEw45mwO1k=">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</latexit><latexit
<latexit
f
Neural Nets
Feedforward
Implicit Generative Models
G(N)
• Learn a function so that the push-forward measure of a well-
known distribution, e.g., a Gaussian, gives the data distribution.
-
-
-
-
Diffusion Models
Flow-based Models
Noise N
Variational Autoencoders
sha1_base64="d6XUVd8yjlOwxLR0JfEw45mwO1k=">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</latexit>
sha1_base64="d6XUVd8yjlOwxLR0JfEw45mwO1k=">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</latexit><latexit
sha1_base64="d6XUVd8yjlOwxLR0JfEw45mwO1k=">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</latexit><latexit
sha1_base64="d6XUVd8yjlOwxLR0JfEw45mwO1k=">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</latexit><latexit
<latexit
f
Generative Adversarial Networks
G(N)
Training Implicit Generative Models
i.e., how to fit the function f mapping noise to data distribution.
• We will cover several modern methods for generative modeling,
You might also like
- Question Bank For Certification Programme of Returning OfficersNo ratings yetQuestion Bank For Certification Programme of Returning Officers77 pages
- Research and Practice in Human Resource ManagementNo ratings yetResearch and Practice in Human Resource Management6 pages
- Biblical Lessons from Grandpa: Preparing the Next GenerationsFrom EverandBiblical Lessons from Grandpa: Preparing the Next GenerationsNo ratings yet
- Special and Different: The Autistic Traveler: Judgment, Redemption, & VictoryFrom EverandSpecial and Different: The Autistic Traveler: Judgment, Redemption, & VictoryNo ratings yet
- What Squirt Teaches Me about Jesus: Kids Learning about Jesus while Playing with FidoFrom EverandWhat Squirt Teaches Me about Jesus: Kids Learning about Jesus while Playing with FidoNo ratings yet
- Extreme Rhyming Poetry: Over 400 Inspirational Poems of Wit, Wisdom, and Humor (Five Books in One)From EverandExtreme Rhyming Poetry: Over 400 Inspirational Poems of Wit, Wisdom, and Humor (Five Books in One)No ratings yet
- Mindfulness Full: Relaxing word search puzzles for adults that will keep your mind calm and positive, 50 brain teasers with more than 600 wordsFrom EverandMindfulness Full: Relaxing word search puzzles for adults that will keep your mind calm and positive, 50 brain teasers with more than 600 wordsNo ratings yet
- Desserts for Senior Living Facilities: (Big Batch Baking for Schools, Churches, and Large Institutions)From EverandDesserts for Senior Living Facilities: (Big Batch Baking for Schools, Churches, and Large Institutions)No ratings yet
- 2021.02.05 USAO Press Release, Over $1.1M in Civil Settlements Reaffirm DOJ's Commitment To Preventing Opioid AbuseNo ratings yet2021.02.05 USAO Press Release, Over $1.1M in Civil Settlements Reaffirm DOJ's Commitment To Preventing Opioid Abuse2 pages
- Petts, Ann - Shapley, Bernard - On Supervision - Psychoanalytic and Jungian Analytic Perspectives-Karnac (2007)100% (1)Petts, Ann - Shapley, Bernard - On Supervision - Psychoanalytic and Jungian Analytic Perspectives-Karnac (2007)266 pages
- McDonald's in A Monopolistic CompetitionNo ratings yetMcDonald's in A Monopolistic Competition23 pages
- BS en ISO 12781-2-2011 - Geometrical Product Specifications (GPS) - Flatness - Part 2 - Specification OperatorsNo ratings yetBS en ISO 12781-2-2011 - Geometrical Product Specifications (GPS) - Flatness - Part 2 - Specification Operators24 pages
- FD Revised 5 - Asf Devastation and Financial Performance of Pork Suppliers in Davao CityNo ratings yetFD Revised 5 - Asf Devastation and Financial Performance of Pork Suppliers in Davao City53 pages
- Using Genetic Algorithms in Process Planning For Job Shop MachiningNo ratings yetUsing Genetic Algorithms in Process Planning For Job Shop Machining12 pages
- Fraud Alert!: "@ril - VC" and "@ril - Sg". TheseNo ratings yetFraud Alert!: "@ril - VC" and "@ril - Sg". These2 pages
- (Communication Electronic Circuits) PrefaceNo ratings yet(Communication Electronic Circuits) Preface2 pages
