|
Daniel@0
|
1 <html>
|
|
Daniel@0
|
2 <head>
|
|
Daniel@0
|
3 <title>
|
|
Daniel@0
|
4 Netlab Reference Manual rbfhess
|
|
Daniel@0
|
5 </title>
|
|
Daniel@0
|
6 </head>
|
|
Daniel@0
|
7 <body>
|
|
Daniel@0
|
8 <H1> rbfhess
|
|
Daniel@0
|
9 </H1>
|
|
Daniel@0
|
10 <h2>
|
|
Daniel@0
|
11 Purpose
|
|
Daniel@0
|
12 </h2>
|
|
Daniel@0
|
13 Evaluate the Hessian matrix for RBF network.
|
|
Daniel@0
|
14
|
|
Daniel@0
|
15 <p><h2>
|
|
Daniel@0
|
16 Synopsis
|
|
Daniel@0
|
17 </h2>
|
|
Daniel@0
|
18 <PRE>
|
|
Daniel@0
|
19 h = rbfhess(net, x, t)
|
|
Daniel@0
|
20 [h, hdata] = rbfhess(net, x, t)
|
|
Daniel@0
|
21 h = rbfhess(net, x, t, hdata)
|
|
Daniel@0
|
22 </PRE>
|
|
Daniel@0
|
23
|
|
Daniel@0
|
24
|
|
Daniel@0
|
25 <p><h2>
|
|
Daniel@0
|
26 Description
|
|
Daniel@0
|
27 </h2>
|
|
Daniel@0
|
28 <CODE>h = rbfhess(net, x, t)</CODE> takes an RBF network data structure <CODE>net</CODE>,
|
|
Daniel@0
|
29 a matrix <CODE>x</CODE> of input values, and a matrix <CODE>t</CODE> of target
|
|
Daniel@0
|
30 values and returns the full Hessian matrix <CODE>h</CODE> corresponding to
|
|
Daniel@0
|
31 the second derivatives of the negative log posterior distribution,
|
|
Daniel@0
|
32 evaluated for the current weight and bias values as defined by
|
|
Daniel@0
|
33 <CODE>net</CODE>. Currently, the implementation only computes the
|
|
Daniel@0
|
34 Hessian for the output layer weights.
|
|
Daniel@0
|
35
|
|
Daniel@0
|
36 <p><CODE>[h, hdata] = rbfhess(net, x, t)</CODE> returns both the Hessian matrix
|
|
Daniel@0
|
37 <CODE>h</CODE> and the contribution <CODE>hdata</CODE> arising from the data dependent
|
|
Daniel@0
|
38 term in the Hessian.
|
|
Daniel@0
|
39
|
|
Daniel@0
|
40 <p><CODE>h = rbfhess(net, x, t, hdata)</CODE> takes a network data structure
|
|
Daniel@0
|
41 <CODE>net</CODE>, a matrix <CODE>x</CODE> of input values, and a matrix <CODE>t</CODE> of
|
|
Daniel@0
|
42 target values, together with the contribution <CODE>hdata</CODE> arising from
|
|
Daniel@0
|
43 the data dependent term in the Hessian, and returns the full Hessian
|
|
Daniel@0
|
44 matrix <CODE>h</CODE> corresponding to the second derivatives of the negative
|
|
Daniel@0
|
45 log posterior distribution. This version saves computation time if
|
|
Daniel@0
|
46 <CODE>hdata</CODE> has already been evaluated for the current weight and bias
|
|
Daniel@0
|
47 values.
|
|
Daniel@0
|
48
|
|
Daniel@0
|
49 <p><h2>
|
|
Daniel@0
|
50 Example
|
|
Daniel@0
|
51 </h2>
|
|
Daniel@0
|
52 For the standard regression framework with a Gaussian conditional
|
|
Daniel@0
|
53 distribution of target values given input values, and a simple
|
|
Daniel@0
|
54 Gaussian prior over weights, the Hessian takes the form
|
|
Daniel@0
|
55 <PRE>
|
|
Daniel@0
|
56
|
|
Daniel@0
|
57 h = beta*hdata + alpha*I
|
|
Daniel@0
|
58 </PRE>
|
|
Daniel@0
|
59
|
|
Daniel@0
|
60
|
|
Daniel@0
|
61 <p><h2>
|
|
Daniel@0
|
62 See Also
|
|
Daniel@0
|
63 </h2>
|
|
Daniel@0
|
64 <CODE><a href="mlphess.htm">mlphess</a></CODE>, <CODE><a href="hesschek.htm">hesschek</a></CODE>, <CODE><a href="evidence.htm">evidence</a></CODE><hr>
|
|
Daniel@0
|
65 <b>Pages:</b>
|
|
Daniel@0
|
66 <a href="index.htm">Index</a>
|
|
Daniel@0
|
67 <hr>
|
|
Daniel@0
|
68 <p>Copyright (c) Ian T Nabney (1996-9)
|
|
Daniel@0
|
69
|
|
Daniel@0
|
70
|
|
Daniel@0
|
71 </body>
|
|
Daniel@0
|
72 </html> |
