Feature Extraction & Classification
Actually, Experiments based on supervised or non-supervised classification are boring task. When i extract a brand-new feature from source signal(such as audio, video, etc), it seems that i've been a genius or a super guy! I can extract every features from signal! I can analyze all natures! ha ha ha!
...but boring task has just started as many of you know. feature extraction from signal dataset (NOT A Signal) takes long time. In some experiments, I had to wait for 3 days for extracting all features from dataset.
Recently in my task on acoustic analysis, I should wait for 2 days. Many of our laboratory colleagues don't know about the classification job, so they bothered me with 'why don't you work in laboratory?' but i cannot resonse like this: 'why not? if you give me more machines to extract features from dataset...'
Recent classification is being faster and faster because of the advances of classifier. As many graduate students know about Support Vector Machine (SVM), classification job has been an instant task. When I should input training data and label set into SVM, I feel 'refreshment.' the endurance is over, and I can see the result over them. In fact, many of lasts lead to disappointment of accuracy.
I think that many researchers are still following this standardized procedure: creating dataset, extracting feature, and training and verifying them. Have you ever felt something special in your experimental or implementing procedures?
...but boring task has just started as many of you know. feature extraction from signal dataset (NOT A Signal) takes long time. In some experiments, I had to wait for 3 days for extracting all features from dataset.
Recently in my task on acoustic analysis, I should wait for 2 days. Many of our laboratory colleagues don't know about the classification job, so they bothered me with 'why don't you work in laboratory?' but i cannot resonse like this: 'why not? if you give me more machines to extract features from dataset...'
Recent classification is being faster and faster because of the advances of classifier. As many graduate students know about Support Vector Machine (SVM), classification job has been an instant task. When I should input training data and label set into SVM, I feel 'refreshment.' the endurance is over, and I can see the result over them. In fact, many of lasts lead to disappointment of accuracy.
I think that many researchers are still following this standardized procedure: creating dataset, extracting feature, and training and verifying them. Have you ever felt something special in your experimental or implementing procedures?
Trackback URL >> http://11471178.net/tt/trackback/1370
-
luapz
2008/06/30 06:30
댓글주소
수정/삭제
댓글쓰기
Good morning Superman.-
AKI
2008/07/02 00:20
댓글주소
수정/삭제
안녕 루압쯔.
-
-
exdeep
2008/06/30 08:07
댓글주소
수정/삭제
댓글쓰기
KIN-
AKI
2008/07/02 00:21
댓글주소
수정/삭제
NIK
-
