<html>
  <head>
    <meta http-equiv="Content-Type" content="text/html; charset=utf-8">
  </head>
  <body text="#000000" bgcolor="#FFFFFF">
    <p>what are semantic genetic algorithm ?</p>
    <p>to my knowledge genetic algorithm lead to poor result except as a
      metaheuristic in optimisation problem <br>
    </p>
    <br>
    <div class="moz-cite-prefix">Le 26/10/2017 à 14:40, Jim O'Flaherty a
      écrit :<br>
    </div>
    <blockquote type="cite"
cite="mid:CAKX5GkiQR=kws=z=vUCjXRGLDMG_GYs9DZWNiy4SPph63FANFw@mail.gmail.com">
      <div dir="auto">When I get time to spend dozens of hours on
        computer go again, I plan to play in Robert's area with semantic
        genetic algorithms. I am an Architect Software Engineer.
        Robert's work will allow me better than starting entirely from
        random in much the same way AlphaGo bootstrapped from the 100K
        of professional games. AG0 then leveraged AlphaGo in knowing an
        architecture that was close enough. My intuition is my approach
        will be something similar in it's evolution.
        <div dir="auto"><br>
        </div>
        <div dir="auto">This is the way we're going to "automate"
          creating provided proofing of human cognition styled computer
          go players to assist humans in a gradient ascent learning
          cycle.</div>
        <div dir="auto"><br>
        </div>
        <div dir="auto">So, Robert, I admire and am encouraged by your
          research for my own computer go projects in this area. Keep
          kicking butt in your unique way. We are in an interesting
          transition in this community. Stick it out. It will be worth
          it long term.</div>
      </div>
      <div class="gmail_extra"><br>
        <div class="gmail_quote">On Oct 26, 2017 4:38 AM, "Petri
          Pitkanen" <<a href="mailto:petri.t.pitkanen@gmail.com"
            target="_blank" moz-do-not-send="true">petri.t.pitkanen@gmail.com</a>>
          wrote:<br type="attribution">
          <blockquote class="gmail_quote" style="margin:0 0 0
            .8ex;border-left:1px #ccc solid;padding-left:1ex">
            <div dir="ltr">Unfortunately there is no proof that you
              principles work better than those form eighties. Nor there
              is any agreement that your pronciples form any improvement
              over the old ones. Yes you are a  far better player than
              me and shows that you are 
              <div>- way better at reading </div>
              <div>- have hugely better go understanding, principles if
                you like</div>
              <div><br>
              </div>
              <div>What is missing that I doubt that you can verbalise
                your go understanding to degree that by applying those
                principles  I could become substantially better player.
                again bulleting</div>
              <div>- My reading skills would not get any better hence
                making much of value any learning moot. Obviously issue
                on me not on your principles</div>
              <div>- your principles are more complex than you
                understand. Much of you know is automated to degree that
                it is subconsciousness information. Transferring that
                information if hard. Usually done by re-playing master
                games looking at problems i.e. training the darn neural
                net in the head</div>
              <div><br>
              </div>
              <div>If you can build Go bot about  KGS 3/4dan strength I
                am more than willing to admit you are right and would
                even consider buying your  books.</div>
              <div><br>
              </div>
              <div>Petri</div>
            </div>
            <div class="gmail_extra"><br>
              <div class="gmail_quote">2017-10-26 6:21 GMT+03:00 Robert
                Jasiek <span dir="ltr"><<a
                    href="mailto:jasiek@snafu.de" target="_blank"
                    moz-do-not-send="true">jasiek@snafu.de</a>></span>:<br>
                <blockquote class="gmail_quote" style="margin:0 0 0
                  .8ex;border-left:1px #ccc solid;padding-left:1ex"><span>On
                    25.10.2017 18:17, Xavier Combelle wrote:<br>
                    <blockquote class="gmail_quote" style="margin:0 0 0
                      .8ex;border-left:1px #ccc solid;padding-left:1ex">
                      exact go theory is full of hole.<br>
                    </blockquote>
                    <br>
                  </span>
                  WRT describing the whole game, yes, this is the
                  current state. Solving go in a mathematical sense is a
                  project for centuries.<span><br>
                    <br>
                    <blockquote class="gmail_quote" style="margin:0 0 0
                      .8ex;border-left:1px #ccc solid;padding-left:1ex">
                      Actually, to my knowledge human can't apply only
                      the exact go theory and<br>
                      play a decent game.<br>
                    </blockquote>
                    <br>
                  </span>
                  Only for certain positions of a) late endgame, b)
                  semeais, c) ko.<span><br>
                    <br>
                    <blockquote class="gmail_quote" style="margin:0 0 0
                      .8ex;border-left:1px #ccc solid;padding-left:1ex">
                      If human can't do that, how it will teach a
                      computer to do it magically ?<br>
                    </blockquote>
                    <br>
                  </span>
                  IIRC, Martin Müller implemented CGT endgames a la
                  Mathematical Go Endgames.<span><br>
                    <br>
                    <blockquote class="gmail_quote" style="margin:0 0 0
                      .8ex;border-left:1px #ccc solid;padding-left:1ex">
                      The reason why (b) had became unpopular is because
                      there is no go theory<br>
                      precise enough to implement it as an algorithm<br>
                    </blockquote>
                    <br>
                  </span>
                  There is quite some theory of the 95% principle kind
                  which might be implemented as approximation. E.g.
                  "Usually, defend your weak important group." can be
                  approximated by approximating "group", "important"
                  (its loss is too large in a quick positional
                  judgement), "weak" (can be killed in two successive
                  moves), "defend" (after the move, cannot be killed in
                  two successive moves), "usually" (always, unless there
                  are several such groups and some must be chosen, say,
                  randomly; the approximation being that the alternative
                  strategy of large scale exchange is discarded).<br>
                  <br>
                  Besides, one must prioritise principles to solve
                  conflicting principles by a higher order principle.<br>
                  <br>
                  IMO, such an expert system combined with tree reading
                  and maybe MCTS to emulate reading used when a
                  principle depends on reading can, with an effort of a
                  few manyears of implementation, already achieve
                  amateur mid dan. Not high dan yet because high dans
                  can choose advanced strategies, such as global
                  exchange, and there are no good enough principles for
                  that yet, which would also consider necessary side
                  conditions related to influence, aji etc. I need to
                  work out such principles during the following years.
                  Currently, the state is that weaker principles have
                  identified the major topics (influence, aji etc.) to
                  be considered in fights but they must be refined to
                  create 95%+ principles.<br>
                  <br>
                  ***<br>
                  <br>
                  In the 80s and 90s, expert systems failed to do better
                  than ca. 5 kyu because principles were only marginally
                  better than 50%. Today, (my) average principles
                  discard the weaker, 50% principles and are ca. 75%.
                  Tomorrow, the 75% principles can be discarded for an
                  average of 95% principles. Expert systems get their
                  chance again! Their major disadvantage remains: great
                  manpower is required for implementation. The advantage
                  is semantical understanding.<span
                    class="m_-1173089658150831574HOEnZb"><font
                      color="#888888"><br>
                      <br>
                      -- <br>
                      robert jasiek</font></span>
                  <div class="m_-1173089658150831574HOEnZb">
                    <div class="m_-1173089658150831574h5"><br>
                      ______________________________<wbr>_________________<br>
                      Computer-go mailing list<br>
                      <a href="mailto:Computer-go@computer-go.org"
                        target="_blank" moz-do-not-send="true">Computer-go@computer-go.org</a><br>
                      <a
                        href="http://computer-go.org/mailman/listinfo/computer-go"
                        rel="noreferrer" target="_blank"
                        moz-do-not-send="true">http://computer-go.org/mailman<wbr>/listinfo/computer-go</a></div>
                  </div>
                </blockquote>
              </div>
              <br>
            </div>
            <br>
            ______________________________<wbr>_________________<br>
            Computer-go mailing list<br>
            <a href="mailto:Computer-go@computer-go.org"
              moz-do-not-send="true">Computer-go@computer-go.org</a><br>
            <a
              href="http://computer-go.org/mailman/listinfo/computer-go"
              rel="noreferrer" target="_blank" moz-do-not-send="true">http://computer-go.org/<wbr>mailman/listinfo/computer-go</a><br>
          </blockquote>
        </div>
      </div>
      <br>
      <fieldset class="mimeAttachmentHeader"></fieldset>
      <br>
      <pre wrap="">_______________________________________________
Computer-go mailing list
<a class="moz-txt-link-abbreviated" href="mailto:Computer-go@computer-go.org">Computer-go@computer-go.org</a>
<a class="moz-txt-link-freetext" href="http://computer-go.org/mailman/listinfo/computer-go">http://computer-go.org/mailman/listinfo/computer-go</a></pre>
    </blockquote>
    <br>
  </body>
</html>