Gem pooling、max pooling、average pooling

gem时average pooling 的一般化
gem 的公式如下:
f ( g ) = [ f 1 ( g ) . . . f k ( g ) . . . f K ( g ) ] T = ( 1 ∣ X k ∣ ∑ x x p k ) 1 p k f^{(g)} = [f_1^{(g)}...f_k^{(g)}...f_K^{(g)}]^T = (\frac{1}{|X_k|}\sum_xx^p{_k})^\frac{1}{p_k} f(g)=[f1(g)...fk(g)...fK(g)]T=(Xk1xxpk)pk1

p k = 1 p_k=1 pk=1时,gem pooling 退化为average pooling,即
f ( a ) = [ f 1 ( a ) . . . f k ( a ) . . . f K ( a ) ] T = 1 ∣ X k ∣ ∑ x x f^{(a)} = [f_1^{(a)}...f_k^{(a)}...f_K^{(a)}]^T = \frac{1}{|X_k|}\sum_xx f(a)=[f1(a)...fk(a)...fK(a)]T=Xk1xx

p k = ∞ p_k=\infty pk=时,gem pooling 退化为max pooling,即
f ( m ) = [ f 1 ( m ) . . . f k ( m ) . . . f K ( m ) ] T = m a x x f^{(m)} = [f_1^{(m)}...f_k^{(m)}...f_K^{(m)}]^T = max x f(m)=[f1(m)...fk(m)...fK(m)]T=maxx

你可能感兴趣的:(深度学习算法,深度学习,神经网络,机器学习)