51. L1
regularization
L2
regularization
Cost function(C0) にL1正規化項を追加
※λ:正規化パラメータ
wが大きい: L1 shrinks the weight much less than L2 does.
wが小さい: L1 shrinks the weight much more than L2 does.
Cost function(C0) にL2正規化項を追加
※λ:正規化パラメータ
→The effect of regularization is to make it so the network
prefer to learn small weight
Wight decay it makes weight smaller lower complexity reduce overfitting
引用例:http://sig.tsg.ne.jp/ml2015/ml/2015/06/29/techniques-of-learning.html
63. Is there a deer in
the image?
Where is the deer
in the image?
Where exactly is the
deer? What pixels?
Which images are similar
to the query image?
Image
Classification
Object
detection
Image
segmentation
Image
Similarity
Similar
image
Query
imageYes
64. Jabil, Printed Circuit Board
Defect Detection
Schneider Electric, Circuit
Breaker Defect Detection
Shell, Counting Grey Stock of
Pipes for Inventory
Management
Land O’Lakes, Sustainability
Farming, Map Labeling
Brillio - Which images are
similar to the query image?
Image
Classification
Object
detection
Image
segmentation
Image
Similarity
Darlie, Retail Store Inspection