[adivs] one of what have been resisting ECs is the problem called needle ... [prip-1] a partial matching rule (r-contiguous) eg 110 111 010 011 as self 000 001 100 101 as non-self r=2 is enought to distinguish them However, it is easy to find a counter example of this. [3] describes such an example as r-contiguous matching nomal 110 100 011 001 anomal 101 111 000 010 in which all detectors when r=2 fails to detect => done ---- here positive sample is non-self and negative sample is self since detectors is desined to detect non-self So these terms are defiend in a sence that - True declaration of positive sample, i.e., non-self declared as non-self => true positive False declaration of positive sample, i.e., self declared as non-self => true positive True declaration of negative sample, i.e., self declared as self => true negative false declaration of negative sample, i.e., non-self declared as self => false negative - under this definitions detection rate is tp/(tp+fn) false alarm rate = fp/(tn+fp) - when we plot DR versus FA, the resulted graph is called Receiver Operating Characgeriristics which reflects a tradeoff between false alarm rate and detection rate => done quotation "We haven't found any mathematics in here yet," he said, before blowing into the pipe. "But I'm sure it exists." If anyone can find the formalisms in an amorphous molten blob it must surely be he. --- from the article by New York Times on 15 February 2005 by M. Wertheim quoting the words from Dr. Erik Demaine, at the Massachusetts Institute of Technology. => done reference: MILA -- Multilevel Immune Learning Algorithm Dipankar Dasgupta, Senhua Yu, and Nivedita Sumi Majumdar GECCO 2003, LNCS 2723, pp. 18394, 2003. => done =============================================================== [cisim] english-sty <> one transfer house or another, would you mind letting us know by posting it on the Pogue feedback boards? => done My column in the paper two weeks ago, though, reversed all expectations. => done footnote Thanks to Mariusz RYBNIK at Paris 12 Univ.~for suggestiong this regarding our problem --- personal communication. => done ===== In this paper we report an on-going investigation into how already proposed methods work on a special situation of what we call an-island-in-a-lake. => Ayara et al. The goal is to create artificial sysems that have the ability to differentiate between self and non-self states where self could be normal while non-self anomalcy ... negative selection allows this self non-self discrimination ... The fact still remains that none of the algorithms is able to resolve all the inherent problems associated with detector generation, thus some tradeoffs have to be considered when choosing an algorithm for generating detectors. The inate immunity which we are born with is the first line of defense if it cannot remove the pathgen then adaptive immune system whicl is acquired during our life time takes over. --- we have two different adaptive immune system. One is by b cells which respond to certain --- both B-cells and T-cells [prip-2] principaly (1) generating candidate detectors randomly; (2) checking them one by one if it matches self patterns; (3) eliminating if it matches self, other with put it in the repartoir This costructs detectors' set which detects non-self. --- Naturarlly this principle results in a highter false alarm