МОВ
Разделение плоскости
⃗
⃗≥�
�
�
⃗
�
Правило
принятия решения
⃗
{⃗� ×�+¿+�≥1 ¿⃗� ×⃗�−+�≤−1
⃗
�
Пусть
для примеров на границе
Асирян
1
МОВ
Расстояние между границами
⃗
� +¿ ¿
⃗
� +¿ − ⃗
�− ¿
⃗
�
����h =¿
⃗
� ,‖⃗
�‖=�
�
⃗
−
Асирян
2
‖⃗
�‖
2
���
≈ ��� ‖⃗
�‖ ≈ ���
2
‖⃗
�‖
2
МОВ
Ограничения
Поиск экстремума
1
L=∑ �� − × ∑ ∑ � � � � � � � � ⃗
�� ×⃗
� � ; ∑ �� � � ⃗
� � × �⃗ +�≥ 0
2 � �
�
�
Асирян
3
Questions
Is there a local maxima problem in SVM?
Is there an overfitting problem in SVM?
In which cases the kernel is necessary?
Асирян
4
Questions
Is there a local maxima problem in SVM?
No
Is there an overfitting problem in SVM?
In which cases the kernel is necessary?
Асирян
5
Questions
Is there a local maxima problem in SVM?
No
Is there an overfitting problem in SVM?
Yes
In which cases the kernel is necessary?
Асирян
6
Questions
Is there a local maxima problem in SVM?
No
Is there an overfitting problem in SVM?
Yes
In which cases the kernel is necessary?
When samples are not linearly separable in
that space.
Асирян
7