home *** CD-ROM | disk | FTP | other *** search
Wrap
(*^ ::[paletteColors = 128; fontset = title, "Helvetica", 24, L1, bold, nohscroll; fontset = subtitle, "Helvetica", 10, L1, italic, nohscroll; fontset = subsubtitle, "Helvetica", 10, L1, nohscroll; fontset = section, "Helvetica", 18, L1, bold, nohscroll; fontset = subsection, "Helvetica", 14, L1, bold, nohscroll; fontset = subsubsection, "Times", 14, L1, bold, nohscroll, blackBox; fontset = text, "Times", 12, L1, nohscroll; fontset = smalltext, "Times", 10, L1, nohscroll; fontset = input, "Courier", 10, L1, nowordwrap; fontset = output, "Courier", 10, L2, nowordwrap; fontset = message, "Helvetica", 10, L1, nowordwrap; fontset = print, "Helvetica", 10, L1, nowordwrap; fontset = info, "Helvetica", 10, L1, nowordwrap; fontset = postscript, "Helvetica", 10, L1, nowordwrap; fontset = name, "Helvetica", 9, L1, italic, nohscroll; fontset = header, "Helvetica", 10, L1, nohscroll; fontset = footer, "Helvetica", 10, L1, nohscroll; fontset = help, "Helvetica", 10, L1, nohscroll; fontset = clipboard, "Helvetica", 10, L1, nohscroll; fontset = completions, "Helvetica", 10, L1, nohscroll; fontset = network, "Courier", 10, L2, nowordwrap; fontset = graphlabel, "Courier", 12, L2, nowordwrap; fontset = special1, "Helvetica", 18, L1, bold, nohscroll; fontset = special2, "Helvetica", 14, L1, bold, nohscroll; fontset = special3, "Times", 14, L1, bold, nohscroll, blackBox; fontset = special4, "Helvetica", 10, L1, nohscroll; fontset = special5, "Helvetica", 10, L1, nohscroll;] :[font = title; inactive; startGroup; noPageBreak; ] Neural Networks :[font = subtitle; inactive; noPageBreak; ] Jaques Vidal, Ph.D. :[font = subsubtitle; inactive; ] Computer Science Dept. and Brain Research Institute 3531 Boelter Hall, UCLA Los Angeles, CA From: Exploring Artificial Neural Networks with Mathematica ;[s] 2:0,0;101,1;155,-1; 2:1,12,9,Helvetica,0,10,0,0,0;1,12,9,Helvetica,2,10,0,0,0; :[font = text; inactive; noPageBreak; ] Mathematica is used to model simple neural networks using the theory of self-organizing maps. One-dimensional distributions are considered. ;[s] 2:0,1;11,0;141,-1; 2:1,13,9,Times,0,12,0,0,0;1,13,9,Times,2,12,0,0,0; :[font = subsection; inactive; startGroup; noPageBreak; ] Kernel Interaction :[font = special3; inactive; ] Animation :[font = text; inactive; noPageBreak; ] This animation shows the lateral interaction of a "kernel" function. An interaction kernel function is a profile of lateral interaction to express the amount of weight modification imposed to the neighbors. :[font = postscript; inactive; PICT; pictureID = 2173; output; pictureWidth = 282; pictureHeight = 174; noPageBreak; ] :[font = text; inactive; endGroup; endGroup; noPageBreak; ] The rotation of a neighbor's weight vector will be toward the weight vector of the winning cell for that particular data vector. The influence will be made to decrease with the distance between that cell and the neighbor. Logically, it should be an even function. We can use the familiar exponential decay for convenience (and analytical tractability). ^*)