Cybenko

記号
I_n := [0,1]^n denotes the n-dimensional unit cube
C(I_n) the space of continuous functions on In
\| f \| := \sup_{x \in I_n} |f(x)| the supremum norm of an f∈C(In)
Th. 1
Let σ be any continuous discriminatory function. Then finite sums of the form
G(x)=\sum_{j=1}^N \alpha_j \sigma(\mathbf{y}_j \cdot \mathbf{x} + \theta_j)
are dense in C(In). In other words, given any f∈C(In) and ε>0, there is a sum, G(x),
of the above form, for which
\sup_{x \in I_n} |G(x)-f(x)|<\epsilon
Th. 2
Let σ be any continuous sigmoidal function. Then finite sums of the form
G(x)=\sum_{j=1}^N \alpha_j \sigma(\mathbf{y}_j \cdot \mathbf{x} + \theta_j)
are dense in C(In). In other words, given any f∈C(In) and ε>0, there is a sum, G(x),
of the above form, for which
\sup_{x \in I_n} |G(x)-f(x)|<\epsilon
Th. 4
Let σ be bounded measurable sigmoidal function. Then finite sums of the form
G(x)=\sum_{j=1}^N \alpha_j \sigma(\mathbf{y}_j \cdot \mathbf{x} + \theta_j)
are dense in L1(In). In other words, given any f∈L1(In) and ε>0, there is a sum, G(x),
of the above form, for which
\|G(x)-f(x)\|_{L^1(I_n)}<\epsilon
g(x)=\sum_{j=1}^N w^{(2)}_j \sigma \left( \sum_{i=1}^D w^{(1)}_{ji} x_i + w^{(1)}_{j0} \right)
x_i \xrightarrow{w^{(1)}_{ji}} a^{(1)}_j \xrightarrow{\sigma^{(1)}} z^{(1)}_{j} \xrightarrow{w^{(2)}_{kj}} a^{(2)}_k \xrightarrow{\sigma^{(2)}} y_k
\sigma(t) := \frac{1}{1 + e^{-t}}
C([0,1]^D),L^1([0,1]^D)
C^{D+1}_0(\mathbb{R}^D) \cap L^\infty (\mathbb{R}^D) \quad \| \cdot \|_{L^2}
\{ \mathbf{x}_n, \mathbf{t}_n\}_{n=1}^N
E(\mathbf{w}) = \frac{1}{2}\sum_{n=1}^N |\mathbf{y}_n(\mathbf{w}) - \mathbf{t}_n|^2
E(\mathbf{w}) = -\sum_{n=1}^N \sum_{k=1}^K t_{nk} \ln y_k(\mathbf{x}_n, \mathbf{w})
\mathbf{g} := \frac{\partial E}{\partial \mathbf{w}}
\mathbf{H} := \frac{\partial^2 E}{\partial \mathbf{w}\partial \mathbf{w}^\mathrm{T}}
\mathbf{w}_{t+1} = \mathbf{w}_{t} - \eta_t \mathbf{g}_t
\mathbf{w}_{t+1} = \mathbf{w}_t - \mathbf{H}^{-1}_t \mathbf{g}_t
\mathbf{w}_{t+1} = \mathbf{w}_t - \eta_t (\mathbf{H}_t + \lambda \mathbf{I})^{-1} \mathbf{g}_t
\mathbf{w}_{t+1} = \mathbf{w}_t - \eta_t \mathbf{G}_t \mathbf{g}_t
\mathbf{d}_t = - \eta_t \mathbf{G}_t \mathbf{g}_t
\mathbf{w}_{t+1} = \mathbf{w}_t - \eta_t \mathbf{G}_t \mathbf{g}_t
g_n(\mathbf{x}) = \frac{1}{n} \sum_{i=1}^n c(\mathbf{a}_i, b_i) \phi_c(\mathbf{a}_i \cdot \mathbf{x} - b_i)
c(\mathbf{a},b) = \sgn(\Re T(\mathbf{a},b)) C_T
C_T := \int |(\mathbf{a},b)| d\mathbf{a} db
最終更新:2009年09月11日 22:44
ツールボックス

下から選んでください:

新しいページを作成する
ヘルプ / FAQ もご覧ください。