0.25
0.5
0.75
1.25
1.5
1.75
2
Kernel methods for structured data
Published on Jan 31, 20171044 Views
Chapter list
Kernel methods for structured data00:00
Outline00:02
Part 102:43
Structured data02:45
Example: Chemoinformatics04:10
Example: Protein function (Borgwardt et al. 2005b)05:25
Example: Sentence classification07:38
Example: sub-community identification (Yanardag et al. 2015)10:04
Structured output learning11:18
Example: Supervised sequence learning12:17
Example: Natural language parsing13:15
Part 213:41
Kernels - 114:01
Kernels - 215:31
Check that your kernel looks reasonable20:14
Advantages of the kernel trick22:27
Example: support vector classification (SVC)25:51
Disadvantages of kernels26:10
Closure properties26:38
Sum and product of kernels27:16
Normalization28:44
Tensor product and direct sum kernels30:38
Convolution (or decomposition) kernels (Haussler 1999)32:04
Example on strings34:12
Example on trees35:05
Co-rooted subtrees (Collins et al. 2001)35:53
Example on graphs37:05
Graphlets, subgraphs38:22
Convolution kernels39:23
Part 340:31
Path kernels40:43
From path kernels to graph kernels42:21
Marginalized graph kernels43:42
Product graphs - 143:44
Product graphs - 246:00
Tottering46:49
Shortest-path graph kernel47:09
Graph isomorphism49:37
Are these two graphs isomorphic?52:10
W-L test: start with all ones52:31
W-L test: propagate52:58
W-L test: recolor53:28
W-L test: propagate again54:22
W-L test: recolor again: isomorphic!54:58
W-L Graph kernel: propagation55:30
W-L Graph kernel: feature vectors (Shervashidze et al. 2011)56:50
W-L Graph kernel: Remarks58:39
W-L Graph kernel: Results (CPU time)58:45
W-L Graph kernel: Results (accuracy)01:00:35
Neighborhood Subgraph Pairwise Distance Kernel - 101:01:08
Neighborhood Subgraph Pairwise Distance Kernel - 201:02:30
Neighborhood Subgraph Pairwise Distance Kernel - 301:03:16
NSPKD Results (CPU time)01:05:45
NSPKD Results (accuracy)01:06:15
Part 401:06:27
kLog01:06:55
Goals01:08:28
Modeling the UW-CSE dataset in kLog - 101:08:32
Modeling the UW-CSE dataset in kLog - 201:09:52
Modeling the UW-CSE dataset in kLog - 301:10:11
Learning from interpretations: predictors and responses01:10:19
Adding background knowledge: Intensional signatures01:11:30
Graphicalization: from interpretations to bipartite graphs01:13:32
Graphicalization in UW-CSE01:14:26
Using graph kernels to construct features01:14:54
Soft matches01:16:35
Supervised learning01:16:41
Viewpoints example01:17:54
A whole kLog script01:19:43
Example: UW-CSE (All information)01:19:44
WebKB: results01:20:54
IMDb: results01:21:37