Butz, M. V., Wilson, S. W.: An algorithmic description of XCS. Journal of Soft Computing, 6 (2002) 144--153.  Home/Search   Document Details and Download   Summary   Related Articles   Check

....the form c 1 ,s 1 , c n ,s n , where c is the condition s range centre and s is the spread from that centre over which the variable is matched by the rule. When a centre with added spread goes outside the defined range it is truncated. All other XCS processing remains as described in [6] except that mutation is done via a random step (range 0.1 x 0.1) cover produces rules centred on the input value with a range of s 0 and subsumption considers the variable ranges. Wilson [20] tested XCSR on a real numbered version of the well known Boolean multiplexer problem. These ....

....as did Wilson s result [20] the increased difficulty. a) b) Figure 7: The effects on population size and GA subsumption on the harder non stationary task for XCSR and sXCSR shown in Figure 5. Similar results were obtained without action set subsumption (Figure 6) Butz and Wilson [6] briefly note that action set subsumption may be detrimental in non stationary tasks due to its reduction in niche diversity. However, the results in Figure 6 suggest that, for the task examined here at least, this is not a significant factor. Analysis of runs in Figure 5 show that a drop in GA ....

Butz, M. and Wilson, S. (2001) An algorithmic description of XCS. In Lanzi, P. L., Stolzmann, W., and S. W. Wilson (Eds.), Advances in Learning Classifier Systems. Third International Workshop (IWLCS-2000), Lecture Notes in Artificial Intelligence (LNAI-1996). Berlin: Springer-Verlag (2001).

....whether it was possible to classify problems into those on which a given algorithm was a winner and those on which it was not a winner. In every case, it was not possible, and therefore the learning task faced by XCS was not a trivial one. 6 THE EXPERIMENTS We used Martin Butz version of XCS [3, 4, 5] available free over the web from the IlliGAL site. We used a single step environment, in which a reward is available at every step, and we de ned a step as packing one bin (LFD was modi ed to pack no more than one bin before returning) The reward earned is proportional to how well lled that ....

Martin V. Butz and Stewart W. Wilson. An Algorithmic Description of XCS. Technical Report 2000.

....whether it was possible to classify problems into those on which a given algorithm was a winner and those on which it was not a winner. In every case, it was not possible, and therefore the learning task faced by XCS was not a trivial one. 6 THE EXPERIMENTS We used Martin Butz version of XCS [3, 4, 5] available free over the web from the IlliGAL site. We used a single step environment, in which a reward is available at every step, and we defined a step as packing one bin (LFD was modified to pack no more than one bin before returning) The reward earned is proportional to how well filled that ....

Martin V. Butz and Stewart W. Wilson. An Algorithmic Description of XCS. Technical Report 2000.

.... of introductory (as well as advanced) material on LCS (see [17] Recent introductory material on some of the currently more popular variations of LCS includes an introduction to Anticipatory Classifier Systems (ACS) 30] an introduction to Fuzzy LCS [6] and introductory material on XCS, e.g. [32, 33, 19, 8]. Given the rapid development of research in XCS perhaps a new comprehensive introductory document will soon be needed. More generally, there is certainly more than enough material to form an introductory text devoted solely to LCS, and, given the renewed interest in LCS in recent years, this may ....

....their own work. 6 Software Alwyn Barry has an impressive collection of LCS software at the LCSWeb [4] including versions of Goldberg s SCS (see [12] for documentation) Riolo s CFS S and CFS C systems (see [28] Grefenstette et al. s SAMUEL (see [11] and several versions of Wilson s XCS (see [8] for documentation of one implementation) ENCORE [13] has a smaller collection of LCS software. Making source code publically available is a great service to the LCS community. Source code provides a level of detail concerning implementation impossible to include in publications. It allows ....

Martin V. Butz and Stewart W. Wilson. An Algorithmic Description of XCS. To appear in [22].

....as opposed to the evolution of one (large) network. SANE [11] is most similar to the work described here, however SANE coevolves individual neurons rather than small networks of neurons as rules. In this paper we implement the new rule representation within XCS, using the specification given in [4]. Results are presented for the single step multiplexer, a multi step maze, and function approximation problems, optimal performance being obtained in all cases. Analysis shows that multiple action rules emerge such that one appropriate rule gives a different action for different inputs. That ....

....the action corresponding to the output node with the highest activation. This matching procedure is repeated for all rules on each cycle. All other system processes are as in XCS, although we do not incorporate subsumption or explicitly maintain at least one entry for each possible action [4]. Rule discovery operates in the same was as usual for XCS with real numbers [16] Hence the mutation operator is altered to adjust gene values using a normal distribution; small changes in weights are more likely than large changes upon satisfaction of the mutation probability (m) The cover ....

Butz, M. & Wilson, S.W. (2001) An Algorithmic Description of XCS. Soft Computing.

....with the classifier; the others are prediction error, fitness, and action set size. Two further associated quantities are the classifier s numerosity and experience. A full description of XCSI may be found in [8] Information on basic XCS is in [6] 7] and in the updated formal description of [2] that XCSI follows most closely. The present work concerns a method for post processing classifier populations already evolved by XCSI. We therefore omit a review of XCSI s mechanisms except for some observations relevant to the post processing. XCSI trains on a dataset like WBC roughly as ....

Martin V. Butz and Stewart W. Wilson. An Algorithmic Description of XCS. In Lanzi et al. [4].

No context found.

Butz, M. V., Wilson, S. W.: An algorithmic description of XCS. Journal of Soft Computing, 6 (2002) 144--153.

No context found.

Martin V. Butz and Stewart W. Wilson. An Algorithmic Description of XCS. To appear in #22#.

No context found.

M. V. Butz and S. W. Wilson. An algorithmic description of XCS. In Lanzi et al. [21], pages 253--272.

No context found.

Martin V. Butz and Stewart W. Wilson. An Algorithmic Description of XCS. In Pier Luca Lanzi, Wolfgang Stolzmann, and Stewart W. Wilson, editors, Advances in Learning Classi#er Systems,number 1996 in LNAI, pages 253#272. Springer# Verlag, 2001.

Online articles have much greater impact   More about CiteSeer.IST   Add search form to your site   Submit documents   Feedback  

CiteSeer.IST - Copyright Penn State and NEC