📄 node18.html
字号:
<!DOCTYPE HTML PUBLIC "-//IETF//DTD HTML 2.0//EN">
<!--Converted with LaTeX2HTML 96.1-h (September 30, 1996) by Nikos Drakos (nikos@cbl.leeds.ac.uk), CBLU, University of Leeds -->
<HTML>
<HEAD>
<TITLE>Computational issues of simulated annealing</TITLE>
<META NAME="description" CONTENT="Computational issues of simulated annealing">
<META NAME="keywords" CONTENT="Surrogates">
<META NAME="resource-type" CONTENT="document">
<META NAME="distribution" CONTENT="global">
<LINK REL=STYLESHEET HREF="Surrogates.css" tppabs="http://www.mpipks-dresden.mpg.de/~tisean/TISEAN_2.0/docs/surropaper/Surrogates.css">
</HEAD>
<BODY bgcolor=#ffffff LANG="EN" >
<A NAME="tex2html255" HREF="node19.html" tppabs="http://www.mpipks-dresden.mpg.de/~tisean/TISEAN_2.0/docs/surropaper/node19.html"><IMG WIDTH=37 HEIGHT=24 ALIGN=BOTTOM ALT="next" SRC="next_motif.gif" tppabs="http://www.mpipks-dresden.mpg.de/~tisean/TISEAN_2.0/docs/surropaper/next_motif.gif"></A> <A NAME="tex2html253" HREF="node16.html" tppabs="http://www.mpipks-dresden.mpg.de/~tisean/TISEAN_2.0/docs/surropaper/node16.html"><IMG WIDTH=26 HEIGHT=24 ALIGN=BOTTOM ALT="up" SRC="up_motif.gif" tppabs="http://www.mpipks-dresden.mpg.de/~tisean/TISEAN_2.0/docs/surropaper/up_motif.gif"></A> <A NAME="tex2html247" HREF="node17.html" tppabs="http://www.mpipks-dresden.mpg.de/~tisean/TISEAN_2.0/docs/surropaper/node17.html"><IMG WIDTH=63 HEIGHT=24 ALIGN=BOTTOM ALT="previous" SRC="previous_motif.gif" tppabs="http://www.mpipks-dresden.mpg.de/~tisean/TISEAN_2.0/docs/surropaper/previous_motif.gif"></A> <BR>
<B> Next:</B> <A NAME="tex2html256" HREF="node19.html" tppabs="http://www.mpipks-dresden.mpg.de/~tisean/TISEAN_2.0/docs/surropaper/node19.html">Example: avoiding periodicity artefacts</A>
<B>Up:</B> <A NAME="tex2html254" HREF="node16.html" tppabs="http://www.mpipks-dresden.mpg.de/~tisean/TISEAN_2.0/docs/surropaper/node16.html">General constrained randomisation</A>
<B> Previous:</B> <A NAME="tex2html248" HREF="node17.html" tppabs="http://www.mpipks-dresden.mpg.de/~tisean/TISEAN_2.0/docs/surropaper/node17.html">Null hypothesesconstraints, and </A>
<BR> <P>
<H2><A NAME="SECTION00052000000000000000">Computational issues of simulated annealing</A></H2>
<P>
Simulated annealing is a very powerful method of combinatorial minimisation in
the presence of many false minima. Simulated annealing has a rich literature,
classical references are Metropolis et al. [<A HREF="node36.html#metro" tppabs="http://www.mpipks-dresden.mpg.de/~tisean/TISEAN_2.0/docs/surropaper/node36.html#metro">36</A>] and
Kirkpatrick [<A HREF="node36.html#kirk" tppabs="http://www.mpipks-dresden.mpg.de/~tisean/TISEAN_2.0/docs/surropaper/node36.html#kirk">37</A>], more recent material can be found for example in
Vidal [<A HREF="node36.html#annealbook" tppabs="http://www.mpipks-dresden.mpg.de/~tisean/TISEAN_2.0/docs/surropaper/node36.html#annealbook">38</A>]. Despite its many successful applications, using
simulated annealing efficiently is still a bit of an art. We will here discuss
some issues we have found worth dealing with in our particular minimisation
problem. Since the detailed behaviour will be different for each cost function,
we can only give some general guidelines.
<P>
<P>
The main idea behind simulated annealing is to interpret the cost function <I>E</I>
as an energy in a thermodynamic system. Minimising the cost function is then
equivalent to finding the ground state of a system. A glassy solid can be
brought close to the energetically optimal state by first heating it and
subsequently cooling it. This procedure is called ``annealing'', hence the name
of the method. If we want to simulate the thermodynamics of this tempering
procedure on a computer, we notice that in thermodynamic equilibrium at some
finite temperature <I>T</I>, system configurations should be visited with a
probability according to the Boltzmann distribution <IMG WIDTH=43 HEIGHT=15 ALIGN=BOTTOM ALT="tex2html_wrap_inline2166" SRC="img97.gif" tppabs="http://www.mpipks-dresden.mpg.de/~tisean/TISEAN_2.0/docs/surropaper/img97.gif"> of the canonical
ensemble. In Monte Carlo simulations, this is achieved by accepting changes of
the configuration with a probability <I>p</I>=1 if the energy is decreased <IMG WIDTH=65 HEIGHT=24 ALIGN=MIDDLE ALT="tex2html_wrap_inline2170" SRC="img98.gif" tppabs="http://www.mpipks-dresden.mpg.de/~tisean/TISEAN_2.0/docs/surropaper/img98.gif"> and <IMG WIDTH=83 HEIGHT=30 ALIGN=MIDDLE ALT="tex2html_wrap_inline2172" SRC="img99.gif" tppabs="http://www.mpipks-dresden.mpg.de/~tisean/TISEAN_2.0/docs/surropaper/img99.gif"> if the energy is increased, <IMG WIDTH=65 HEIGHT=24 ALIGN=MIDDLE ALT="tex2html_wrap_inline2174" SRC="img100.gif" tppabs="http://www.mpipks-dresden.mpg.de/~tisean/TISEAN_2.0/docs/surropaper/img100.gif">. This selection rule is often referred to as the <EM>Metropolis step</EM>. In
a minimisation problem, the temperature is the parameter in the Boltzmann
distribution that sets its width. In particular, it determines the probability
to go ``up hill'', which is important if we need to get out of false minima.
<P>
In order to anneal the system to the ground state of minimal ``energy'', that
is, the minimum of the cost function, we want to first ``melt'' the system at a
high temperature <I>T</I>, and then decrease <I>T</I> slowly, allowing the system to be
close to thermodynamic equilibrium at each stage. If the changes to the
configuration we allow to be made connect all possible states of the system,
the updating algorithm is called <EM>ergodic</EM>. Although some general rigorous
convergence results are available, in practical applications of simulated
annealing some problem-specific choices have to be made. In particular, apart
from the constraints and the cost function, one has to specify a method of
updating the configurations and a schedule for lowering the temperature. In the
following, we will discuss each of these issues.
<P>
Concerning the choice of cost function, we have already mentioned that there is
a large degeneracy in that many cost functions have an absolute minimum
whenever a given set of constraints if fulfilled. The convergence properties
can depend dramatically on the choice of cost function. Unfortunately, this
dependence seems to be so complicated that it is impossible even to discuss the
main behaviour in some generality. In particular, the weights <IMG WIDTH=15 HEIGHT=14 ALIGN=MIDDLE ALT="tex2html_wrap_inline2180" SRC="img101.gif" tppabs="http://www.mpipks-dresden.mpg.de/~tisean/TISEAN_2.0/docs/surropaper/img101.gif"> in
Eq.(<A HREF="node17.html#eqE" tppabs="http://www.mpipks-dresden.mpg.de/~tisean/TISEAN_2.0/docs/surropaper/node17.html#eqE">22</A>) are sometimes difficult to choose. Heuristically, we would like
to reflect changes in the <I>I</I> different constraints about equally, provided the
constraints are independent. Since their scale is not at all set by
Eq.(<A HREF="node17.html#eqF" tppabs="http://www.mpipks-dresden.mpg.de/~tisean/TISEAN_2.0/docs/surropaper/node17.html#eqF">21</A>), we can use the <IMG WIDTH=15 HEIGHT=14 ALIGN=MIDDLE ALT="tex2html_wrap_inline2180" SRC="img101.gif" tppabs="http://www.mpipks-dresden.mpg.de/~tisean/TISEAN_2.0/docs/surropaper/img101.gif"> for this purpose. Whenever we have some
information about which kind of deviation would be particularly problematic
with a given test statistic, we can give it a stronger weight. Often, the
shortest lags of the autocorrelation function are of crucial importance, whence
we tend to weight autocorrelations by <IMG WIDTH=23 HEIGHT=24 ALIGN=MIDDLE ALT="tex2html_wrap_inline2186" SRC="img102.gif" tppabs="http://www.mpipks-dresden.mpg.de/~tisean/TISEAN_2.0/docs/surropaper/img102.gif"> when they occur in sums. Also,
the <IMG WIDTH=31 HEIGHT=24 ALIGN=MIDDLE ALT="tex2html_wrap_inline2060" SRC="img57.gif" tppabs="http://www.mpipks-dresden.mpg.de/~tisean/TISEAN_2.0/docs/surropaper/img57.gif"> with larger <IMG WIDTH=8 HEIGHT=7 ALIGN=BOTTOM ALT="tex2html_wrap_inline2190" SRC="img103.gif" tppabs="http://www.mpipks-dresden.mpg.de/~tisean/TISEAN_2.0/docs/surropaper/img103.gif"> are increasingly ill-determined due to the
fewer data points entering the sums. As an extreme example, <IMG WIDTH=135 HEIGHT=24 ALIGN=MIDDLE ALT="tex2html_wrap_inline2192" SRC="img104.gif" tppabs="http://www.mpipks-dresden.mpg.de/~tisean/TISEAN_2.0/docs/surropaper/img104.gif">
shows huge fluctuations due to the lack of self-averaging. Finally, there are
many more <IMG WIDTH=31 HEIGHT=24 ALIGN=MIDDLE ALT="tex2html_wrap_inline2060" SRC="img57.gif" tppabs="http://www.mpipks-dresden.mpg.de/~tisean/TISEAN_2.0/docs/surropaper/img57.gif"> with larger <IMG WIDTH=8 HEIGHT=7 ALIGN=BOTTOM ALT="tex2html_wrap_inline2190" SRC="img103.gif" tppabs="http://www.mpipks-dresden.mpg.de/~tisean/TISEAN_2.0/docs/surropaper/img103.gif"> than at the crucial short lags.
<P>
<blockquote><A NAME="957"> </A><IMG WIDTH=330 HEIGHT=321 ALIGN=BOTTOM ALT="figure1074" SRC="img96.gif" tppabs="http://www.mpipks-dresden.mpg.de/~tisean/TISEAN_2.0/docs/surropaper/img96.gif"><BR>
<STRONG>Figure:</STRONG> <A NAME="figupdate"> </A>
Building up correlations by pairwise permutation. Suppose we want to
generate the strong anti-correlation present in the data (upper trace) by
minimising <IMG WIDTH=126 HEIGHT=29 ALIGN=MIDDLE ALT="tex2html_wrap_inline2144" SRC="img95.gif" tppabs="http://www.mpipks-dresden.mpg.de/~tisean/TISEAN_2.0/docs/surropaper/img95.gif">. The annealing started with a random
permutation (middle trace, <I>E</I>=1.129). At a given intermediate state (lower
trace, <I>E</I>=0.256), exchanging the points <I>a</I> and <I>b</I> increases the
cost to <I>E</I>=0.2744 while exchanging <I>c</I> and <I>d</I> creates negative correlation
and reduces the cost to <I>E</I>=0.002.<BR>
</blockquote>
<p>
A way to efficiently reach all permutations by small individual changes is by
exchanging randomly chosen (not necessarily close-by) pairs. How the
interchange of two points can affect the current cost is illustrated
schematically in Fig. <A HREF="node18.html#figupdate" tppabs="http://www.mpipks-dresden.mpg.de/~tisean/TISEAN_2.0/docs/surropaper/node18.html#figupdate">10</A>. Optimising the code that computes and
updates the cost function is essential since we need its current value at each
annealing step -- which are expected to be many. Very often, an exchange of
two points is reflected in a rather simple update of the cost function. For
example, computing <IMG WIDTH=31 HEIGHT=24 ALIGN=MIDDLE ALT="tex2html_wrap_inline2060" SRC="img57.gif" tppabs="http://www.mpipks-dresden.mpg.de/~tisean/TISEAN_2.0/docs/surropaper/img57.gif"> for a single lag <IMG WIDTH=8 HEIGHT=7 ALIGN=BOTTOM ALT="tex2html_wrap_inline2190" SRC="img103.gif" tppabs="http://www.mpipks-dresden.mpg.de/~tisean/TISEAN_2.0/docs/surropaper/img103.gif"> involves <I>O</I>(<I>N</I>)
multiplications. Updating <IMG WIDTH=31 HEIGHT=24 ALIGN=MIDDLE ALT="tex2html_wrap_inline2060" SRC="img57.gif" tppabs="http://www.mpipks-dresden.mpg.de/~tisean/TISEAN_2.0/docs/surropaper/img57.gif"> upon the exchange of two points <I>i</I><<I>j</I> only
requires the replacement of the terms <IMG WIDTH=40 HEIGHT=12 ALIGN=MIDDLE ALT="tex2html_wrap_inline2208" SRC="img105.gif" tppabs="http://www.mpipks-dresden.mpg.de/~tisean/TISEAN_2.0/docs/surropaper/img105.gif">, <IMG WIDTH=40 HEIGHT=15 ALIGN=MIDDLE ALT="tex2html_wrap_inline2210" SRC="img106.gif" tppabs="http://www.mpipks-dresden.mpg.de/~tisean/TISEAN_2.0/docs/surropaper/img106.gif">,
<IMG WIDTH=43 HEIGHT=15 ALIGN=MIDDLE ALT="tex2html_wrap_inline2212" SRC="img107.gif" tppabs="http://www.mpipks-dresden.mpg.de/~tisean/TISEAN_2.0/docs/surropaper/img107.gif">, and <IMG WIDTH=42 HEIGHT=15 ALIGN=MIDDLE ALT="tex2html_wrap_inline2214" SRC="img108.gif" tppabs="http://www.mpipks-dresden.mpg.de/~tisean/TISEAN_2.0/docs/surropaper/img108.gif"> in the sum. Note that cheap updates are a
source of potential mistakes (e.g. avoid subtracting terms twice in the case
that <IMG WIDTH=61 HEIGHT=24 ALIGN=MIDDLE ALT="tex2html_wrap_inline2216" SRC="img109.gif" tppabs="http://www.mpipks-dresden.mpg.de/~tisean/TISEAN_2.0/docs/surropaper/img109.gif">) but also of roundoff errors. To ensure that the assumed cost
is always equal to the actual cost, code carefully and monitor roundoff by
computing a fresh cost function occasionally.
<P>
Further speed-up can be achieved in two ways. Often, not all the terms in a
cost function have to be added up until it is clear that the resulting change
goes up hill by an amount that will lead to a rejection of the exchange.
Also, pairs to be exchanged can be selected closer in magnitude at low
temperatures because large changes are very likely to increase the cost.
<P>
Many cooling schemes have been discussed in the
literature [<A HREF="node36.html#annealbook" tppabs="http://www.mpipks-dresden.mpg.de/~tisean/TISEAN_2.0/docs/surropaper/node36.html#annealbook">38</A>]. We use an exponential scheme in our work. We
will give details on the - admittedly largely ad hoc -- choices that have
been made in the TISEAN implementation in Appendix <A HREF="node32.html#appA" tppabs="http://www.mpipks-dresden.mpg.de/~tisean/TISEAN_2.0/docs/surropaper/node32.html#appA">A</A>. We found it
convenient to have a scheme available that automatically adjusts parameters
until a given accuracy is reached. This can be done by cooling at a certain
rate until we are stuck (no more accepted changes). If the cost is not low
enough yet, we melt the system again and cool at a slower rate.
<P>
<HR><A NAME="tex2html255" HREF="node19.html" tppabs="http://www.mpipks-dresden.mpg.de/~tisean/TISEAN_2.0/docs/surropaper/node19.html"><IMG WIDTH=37 HEIGHT=24 ALIGN=BOTTOM ALT="next" SRC="next_motif.gif" tppabs="http://www.mpipks-dresden.mpg.de/~tisean/TISEAN_2.0/docs/surropaper/next_motif.gif"></A> <A NAME="tex2html253" HREF="node16.html" tppabs="http://www.mpipks-dresden.mpg.de/~tisean/TISEAN_2.0/docs/surropaper/node16.html"><IMG WIDTH=26 HEIGHT=24 ALIGN=BOTTOM ALT="up" SRC="up_motif.gif" tppabs="http://www.mpipks-dresden.mpg.de/~tisean/TISEAN_2.0/docs/surropaper/up_motif.gif"></A> <A NAME="tex2html247" HREF="node17.html" tppabs="http://www.mpipks-dresden.mpg.de/~tisean/TISEAN_2.0/docs/surropaper/node17.html"><IMG WIDTH=63 HEIGHT=24 ALIGN=BOTTOM ALT="previous" SRC="previous_motif.gif" tppabs="http://www.mpipks-dresden.mpg.de/~tisean/TISEAN_2.0/docs/surropaper/previous_motif.gif"></A> <BR>
<B> Next:</B> <A NAME="tex2html256" HREF="node19.html" tppabs="http://www.mpipks-dresden.mpg.de/~tisean/TISEAN_2.0/docs/surropaper/node19.html">Example: avoiding periodicity artefacts</A>
<B>Up:</B> <A NAME="tex2html254" HREF="node16.html" tppabs="http://www.mpipks-dresden.mpg.de/~tisean/TISEAN_2.0/docs/surropaper/node16.html">General constrained randomisation</A>
<B> Previous:</B> <A NAME="tex2html248" HREF="node17.html" tppabs="http://www.mpipks-dresden.mpg.de/~tisean/TISEAN_2.0/docs/surropaper/node17.html">Null hypothesesconstraints, and </A>
<P><ADDRESS>
<I>Thomas Schreiber <BR>
Mon Aug 30 17:31:48 CEST 1999</I>
</ADDRESS>
</BODY>
</HTML>
⌨️ 快捷键说明
复制代码
Ctrl + C
搜索代码
Ctrl + F
全屏模式
F11
切换主题
Ctrl + Shift + D
显示快捷键
?
增大字号
Ctrl + =
减小字号
Ctrl + -