📄 http:^^www-cse.ucsd.edu^users^rik^section3_1.html
字号:
Date: Tue, 14 Jan 1997 23:50:12 GMTServer: NCSA/1.4.2Content-type: text/htmlLast-modified: Tue, 04 Apr 1995 10:30:20 GMTContent-length: 2836<HEAD><TITLE> Introduction</TITLE></HEAD><BODY><P> <HR> <!WA0><A NAME=tex2html24 HREF="http://www-cse.ucsd.edu/users/rik/section3_2.html"><!WA1><IMG ALIGN=MIDDLE SRC="http://www-cse.ucsd.edu/icons/latex2html/next_motif.gif"></A> <!WA2><A NAME=tex2html22 HREF="http://www-cse.ucsd.edu/users/rik/research-stmnt.html"><!WA3><IMG ALIGN=MIDDLE SRC="http://www-cse.ucsd.edu/icons/latex2html/up_motif.gif"></A> <!WA4><A NAME=tex2html16 HREF="http://www-cse.ucsd.edu/users/rik/research-stmnt.html"><!WA5><IMG ALIGN=MIDDLE SRC="http://www-cse.ucsd.edu/icons/latex2html/previous_motif.gif"></A> <BR><B> Next:</B> <!WA6><A NAME=tex2html25 HREF="http://www-cse.ucsd.edu/users/rik/section3_2.html"> Adaptive information retrieval</A><B>Up:</B> <!WA7><A NAME=tex2html23 HREF="http://www-cse.ucsd.edu/users/rik/research-stmnt.html">Belew: Research Statement</A><B> Previous:</B> <!WA8><A NAME=tex2html17 HREF="http://www-cse.ucsd.edu/users/rik/research-stmnt.html">Belew: Research Statement</A><HR> <P><H1><A NAME=SECTION0001000000000000000> Introduction</A></H1><P>My research focus is a characterization of adaptive knowledgerepresentations. Issues of representation have always played acentral role in artificial intelligence (AI), as well as in computerscience and theories of mind more generally. But I would argue thatmost of this work has (implicitly or explicitly) assumed that therepresentational language is wielded <em>manually</em>, by humansencoding an explicit characterization of what they believe to be trueof the world. Philosophical difficulties aside, some modern machinelearning techniques are capable of <em>automatically</em> developingelaborate representations of the world. A central result of themathematical theory of induction is that the selection of anappropriate language for representing learned concepts is absolutelycritical to their identification. It is therefore appropriate toreconsider basic notions of what makes for good knowledgerepresentation, with constraints imposed by the learning processconsidered <em>sine qua non</em>, along with those (expressive adequacy,valid inference, etc.) more typically considered by AI.<P>I have found it productive to pursue this general interest throughtwo more specific research projects. The first of these usesconnectionist (neural) networks as a representation for theinformation retrieval (IR) problem. This construction allows an IRsystem to learn a more effective indexing representation of free-textdocuments as a simple by-product of the browsing behaviors of itsusers. Second and more recently, I have investigated GeneticAlgorithm (GAs), particularly interactions between neural networks andthe GA, both as algorithmic techniques and as models of naturalphenomena (learning and evolution, resp.). I have found that my workin these two areas allows a ``stereoscopic'' view, encompassing bothlow-level biological constraints and high-level cultural issues, thatare at the heart of modern AI and cognitive science.<P><HR></BODY><P><ADDRESS>rik@cs.ucsd.edu</ADDRESS>⌨️ 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 + -