Mercurial > cortex

comparison org/ullman.org @ 380:2d0afb231081

Find changesets by keywords (author, files, the commit message), revision number or hash, or revset expression.
spellcheck
author rlm
date Wed, 10 Apr 2013 16:49:05 -0400
parents f1b8727360fb
children
comparison
equal deleted inserted replaced
379:f1b8727360fb 380:2d0afb231081
11 * Ullman 11 * Ullman
12 12
13 Actual code reuse! 13 Actual code reuse!
14 14
15 precision = fraction of retrieved instances that are relevant 15 precision = fraction of retrieved instances that are relevant
16 (true-postives/(true-positives+false-positives)) 16 (true-positives/(true-positives+false-positives))
17 17
18 recall = fraction of relevant instances that are retrieved 18 recall = fraction of relevant instances that are retrieved
19 (true-positives/total-in-class) 19 (true-positives/total-in-class)
20 20
21 cross-validation = train the model on two different sets to prevent 21 cross-validation = train the model on two different sets to prevent
22 overfitting. 22 overfitting, and confirm that you have enough training samples.
23 23
24 nifty, relevant, realistic ideas 24 nifty, relevant, realistic ideas
25 He doesn't confine himself to unplasaubile assumptions 25 He doesn't confine himself to implausible assumptions
26 26
27 ** Our Reading 27 ** Our Reading
28 28
29 *** 2002 Visual features of intermediate complexity and their use in classification 29 *** 2002 Visual features of intermediate complexity and their use in classification
30 30
31 31
32 32
33
34 Viola's PhD thesis has a good introduction to entropy and mutual
35 information
33 36
34 ** Getting around the dumb "fixed training set" methods 37 ** Getting around the dumb "fixed training set" methods
35 38
36 *** 2006 Learning to classify by ongoing feature selection 39 *** 2006 Learning to classify by ongoing feature selection
37 40
38 Brings in the most informative features of a class, based on 41 Brings in the most informative features of a class, based on
39 mutual information between that feature and all the examples 42 mutual information between that feature and all the examples
40 encountered so far. To bound the running time, he uses only a 43 encountered so far. To bound the running time, he uses only a
41 fixed number of the most recent examples. He uses a replacement 44 fixed number of the most recent examples. He uses a replacement
42 strategy to tell whether a new feature is better than one of the 45 strategy to tell whether a new feature is better than one of the
43 corrent features. 46 current features.
44 47
45 *** 2009 Learning model complexity in an online environment 48 *** 2009 Learning model complexity in an online environment
46 49
47 Sort of like the heirichal baysean models of Tennanbaum, this 50 Sort of like the hierarchical Bayesan models of Tennanbaum, this
48 system makes the model more and more complicated as it gets more 51 system makes the model more and more complicated as it gets more
49 and more training data. It does this by using two systems in 52 and more training data. It does this by using two systems in
50 parallell and then whenever the more complex one seems to be 53 parallel and then whenever the more complex one seems to be
51 needed by the data, the less complex one is thrown out, and an 54 needed by the data, the less complex one is thrown out, and an
52 even more complex model is initialized in its place. 55 even more complex model is initialized in its place.
53 56
54 He uses a SVM with polynominal kernels of varying complexity. He 57 He uses a SVM with polynomial kernels of varying complexity. He
55 gets good perfoemance on a handwriting classfication using a large 58 gets good performance on a handwriting classification using a large
56 range of training samples, since his model changes complexity 59 range of training samples, since his model changes complexity
57 depending on the number of training samples. The simpler models do 60 depending on the number of training samples. The simpler models do
58 better with few training points, and the more complex ones do 61 better with few training points, and the more complex ones do
59 better with many training points. 62 better with many training points.
60 63
70 [[../images/viola-parzen-1.png]] 73 [[../images/viola-parzen-1.png]]
71 [[../images/viola-parzen-2.png]] 74 [[../images/viola-parzen-2.png]]
72 75
73 *** 2010 The chains model for detecting parts by their context 76 *** 2010 The chains model for detecting parts by their context
74 77
75 Like the constelation method for rigid objects, but extended to 78 Like the constellation method for rigid objects, but extended to
76 non-rigid objects as well. 79 non-rigid objects as well.
77 80
78 Allows you to build a hand detector from a face detector. This is 81 Allows you to build a hand detector from a face detector. This is
79 usefull because hands might be only a few pixels, and very 82 useful because hands might be only a few pixels, and very
80 ambiguous in an image, but if you are expecting them at the end of 83 ambiguous in an image, but if you are expecting them at the end of
81 an arm, then they become easier to find. 84 an arm, then they become easier to find.
82 85
83 They make chains by using spatial proximity of features. That way, 86 They make chains by using spatial proximity of features. That way,
84 a hand can be idntified by chaining back from the head. If there 87 a hand can be identified by chaining back from the head. If there
85 is a good chain to the head, then it is more likely that there is 88 is a good chain to the head, then it is more likely that there is
86 a hand than if there isn't. Since there is some give in the 89 a hand than if there isn't. Since there is some give in the
87 proximity detection, the system can accomodate new poses that it 90 proximity detection, the system can accommodate new poses that it
88 has never seen before. 91 has never seen before.
89 92
90 Does not use any motion information. 93 Does not use any motion information.
91 94
92 *** 2005 A Hierarchical Non-Parametric Method for Capturing Non-Rigid Deformations 95 *** 2005 A Hierarchical Non-Parametric Method for Capturing Non-Rigid Deformations
96 Goal is to match images, as in SIFT, but this time the images can 99 Goal is to match images, as in SIFT, but this time the images can
97 be subject to non rigid transformations. They do this by finding 100 be subject to non rigid transformations. They do this by finding
98 small patches that look the same, then building up bigger 101 small patches that look the same, then building up bigger
99 patches. They get a tree of patches that describes each image, and 102 patches. They get a tree of patches that describes each image, and
100 find the edit distance between each tree. Editing operations 103 find the edit distance between each tree. Editing operations
101 involve a coherent shift of features, so they can accomodate local 104 involve a coherent shift of features, so they can accommodate local
102 shifts of patches in any direction. They get some cool results 105 shifts of patches in any direction. They get some cool results
103 over just straight correlation. Basically, they made an image 106 over just straight correlation. Basically, they made an image
104 comparor that is resistant to multiple independent deformations. 107 comparator that is resistant to multiple independent deformations.
105 108
106 !important small regions are treated the same as nonimportant 109 !important small regions are treated the same as unimportant
107 small regions 110 small regions
108 111
109 !no conception of shape 112 !no conception of shape
110 113
111 quote: 114 quote:
116 performed one on top of the other. 119 performed one on top of the other.
117 120
118 *** 2006 Satellite Features for the Classification of Visually Similar Classes 121 *** 2006 Satellite Features for the Classification of Visually Similar Classes
119 122
120 Finds features that can distinguish subclasses of a class, by 123 Finds features that can distinguish subclasses of a class, by
121 first finding a rigid set of anghor features that are common to 124 first finding a rigid set of anchor features that are common to
122 both subclasses, then finding distinguishing features relative to 125 both subclasses, then finding distinguishing features relative to
123 those subfeatures. They keep things rigid because the satellite 126 those subfeatures. They keep things rigid because the satellite
124 features don't have much information in and of themselves, and are 127 features don't have much information in and of themselves, and are
125 only informative relative to other features. 128 only informative relative to other features.
126 129
127 *** 2005 Learning a novel class from a single example by cross-generalization. 130 *** 2005 Learning a novel class from a single example by cross-generalization.
128 131
129 Let's you use a vast visual experience to generate a classifier 132 Let's you use a vast visual experience to generate a classifier
130 for a novel class by generating synthetic examples by replaceing 133 for a novel class by generating synthetic examples by replacing
131 features from the single example with features from similiar 134 features from the single example with features from similar
132 classes. 135 classes.
133 136
134 quote: feature F is likely to be useful for class C if a similar 137 quote: feature F is likely to be useful for class C if a similar
135 feature F proved effective for a similar class C in the past. 138 feature F proved effective for a similar class C in the past.
136 139
137 Allows you to trasfer the "gestalt" of a similiar class to a new 140 Allows you to transfer the "gestalt" of a similar class to a new
138 class, by adapting all the features of the learned class that have 141 class, by adapting all the features of the learned class that have
139 correspondance to the new class. 142 correspondence to the new class.
140 143
141 *** 2007 Semantic Hierarchies for Recognizing Objects and Parts 144 *** 2007 Semantic Hierarchies for Recognizing Objects and Parts
142 145
143 Better learning of complex objects like faces by learning each 146 Better learning of complex objects like faces by learning each
144 piece (like nose, mouth, eye, etc) separately, then making sure 147 piece (like nose, mouth, eye, etc) separately, then making sure
145 that the features are in plausable positions. 148 that the features are in plausible positions.