Mentales habitudes - Tag - machine learning2015-10-19T19:25:25+02:00nojhanurn:md5:12147DotclearHybridization : estimation of distribution as a meta-model filter generator for metaheuristics ?urn:md5:4137438ac93f0a4807a99223b387a73f2007-07-27T00:00:00+02:00nojhanDiversestimation of distributionmachine learning <p>An interesting idea is to use meta-model (a priori representation of the
problem) as a filter to bias the sample produced by metaheuristics. This
approach seems especially promising for engineering problem, where computing
the objective function is very expensive.</p>
<p>One simple form of meta-model is a probability density function,
approximating the shape of the objective function. This PDF could thus be used
to filter out bad points <em>before</em> evaluation.</p>
<p>Why, then, do not directly use EDA to generate the sample ? Because one can
imagine that the problem shape is not well known, and that using a complex PDF
is impossible (too expensive to compute, for example). Then, using a classical
indirect metaheuristic (let say an evolutionary algorithm) should be preferable
(computationnaly inexpensive) for the sample generation. If one know a good
approximation to use for the distribution of the EDA (not too computationnaly
expensive), one can imagine using the best part of the two worlds.</p>
<p>An example could be a problem with real variable : using an EDA with a
multi-variate normal distribution is computationnaly expensive (due to the
estimation of the co-variance, mainly), and using a mixture of gaussian kernels
makes difficult to have an <em>a priori</em> on the problem. Thus, why not
using a indirect metaheuristic to handle the sample generation, and use a
meta-model which parameters are estimated from the previous sample, according
to a chosen distribution ?</p>
<p>One more hybridization to try...</p>Random and learningurn:md5:6925d6670a20503fbc14021221a840282006-12-21T00:00:00+01:00nojhanDiversmachine learning <p>The <em><a href="http://hunch.net/" hreflang="en">Machine Learning
(Theory)</a></em> blog, by <a href="http://hunch.net/~jl/" hreflang="en">John
Langford</a>, has a very intersting post on "<a href="http://hunch.net/wp-trackback.php?p=239" hreflang="en">Explicit Randomization
in Learning algorithms</a>".</p>
<p>The post and its comments are talking about machine-learning, but can
largely be applied to metaheuristics. The page is listing several reason for
using randomization, from which some are of special intersts for
metaheuristics:</p>
<ol>
<li>symmetry breaking as a way to make decision, which is of great importance
for metaheuristics, which must learn and choose where are the "promising
regions";</li>
<li>overfit avoidance, which is related to the intensification/diversification
balance problem;</li>
<li>adversary defeating and bias suppression, which can be interpreted as
trying to design a true <em>meta</em>-heuristic (i.e. that can be applied on
several problems without major changes).</li>
</ol>
<p>Of course, it should be possible to design a completely deterministic
algorithm that takes decisions, achieve a correct i/d balance and can tackle
all problems... Even if this force to integrate the problems themselves in the
algorithm, it <em>should</em> be possible. The drawback is that it is
computationally intractable.</p>
<p>In fact, metaheuristics (and, as far as I understand, machine-learning
algorithms) are located somewhere between random search algorithms and
deterministic ones. The compromise between these two tendencies is dependent of
the problem and of the offered computational effort.</p>Metaheuristics and machine-learningurn:md5:6a9a108df90979ccc7d46e7b1cf8dcd62006-12-19T00:00:00+01:00nojhanDiversevolutionary computationmachine learningsimulated annealing <p>Metaheuristics and machine-learning algorithms shares a large number of
characteristics, like stochastic processes, manipulaton of probability density
functions, etc.</p>
<p>One of the interesting evolution of the research on metaheuristics these
years is the increasing bridge-building with machine-learning. I see at least
two interesting pathways: the use of metaheuristics in machine-learning and the
use of machine-learning in metaheuristics.</p>
<p>The first point is not really new, machine-learning heavily use
optimization, and it was natural to try stochastic algorithms where local
search or exact algorithms failed. Nevertheless, there is now a sufficient
litterature to organize some special sessions in some symposium. For 2007,
there will be a <em><a href="http://seal.tst.adfa.edu.au/~alar/gbml2007/" hreflang="en">special session on Genetics-Based Machine Learning</a></em> at
CEC, and a <em><a href="http://www.sigevo.org/gecco-2007/organizers-tracks.html" hreflang="en">track
on Genetics-Based Machine Learning and Learning Classifier Systems</a></em> at
GECCO. These events are centered around "genetic" algortihm (see the posts on
the IlliGAL blog : <a href="http://www-illigal.ge.uiuc.edu/system/components/com_jd-wp/wp-trackback.php?p=745" hreflang="en">1</a>, <a href="http://www-illigal.ge.uiuc.edu/system/components/com_jd-wp/wp-trackback.php?p=746" hreflang="en">2</a>), despite the fact that there are several papers using
other metaheuritics, like simulated annealing, but this is a common drawback,
and does not affect the interest of the subject.</p>
<p>The second point is less exploited, but I find it of great interest. A
simple example of what can be done with machine-learning inside metaheuristic
can be shown with estimation of distribution algorithms. In these
metaheuristics, a probability density function is used to explicitely build a
new sample of the objective function (a "population", in the evolutionary
computation terminology) at each iteration. It is then crucial to build a
probability density function that is related to the structure of the objective
function (the "fitness landscape"). There, it should be really interesting to
build the model of the pdf itself from a selected sample, using a
machine-learning algorithm. There is some interesting papers talking about
that.</p>
<p>If you mix these approaches with the problem of estimating a Boltzmann
distribution (the basis of simulated annealing), you should have an awesome
research field...</p>