Xref: utzoo comp.sys.next:10401 comp.ai.neural-nets:2568 Path: utzoo!utgpu!news-server.csri.toronto.edu!cs.utexas.edu!uunet!camex!circus!geoff From: geoff@circus.camex.com (Geoffrey Knauth) Newsgroups: comp.sys.next,comp.ai.neural-nets Subject: balancing seal (or boat) Summary: the balancing seal example gave me another idea Keywords: neural rowing Message-ID: <1685@camex.COM> Date: 5 Dec 90 18:16:59 GMT References: <4385@idunno.Princeton.EDU> <1990Dec2.113653.5222@engin.umich.edu> Sender: news@Camex.COM Reply-To: geoff@circus.UUCP (Geoffrey Knauth) Distribution: na Organization: Camex Inc., Boston, MA Lines: 23 The Balancing Seal (neural net) example not only sparked my interest in neural nets, but it made me wonder if my NeXT can help rowing coaches understand how they can do a better job. Immediately as I watched the seal learn, I realized that scullers seem to look and improve the same way as they learn to balance the boat. At Boston Rowing Center, where I am a coxswain, coaches prepare elite rowers and candidates for World Championships and Olympics. Part of the mystique of coaching is knowing how much information the rower needs in order to improve most rapidly, among other things, technique and balance. I am particularly interested to find out why after 600-700 trials, the seal begins to show much better control, and why, "because of the nature of the neural network learning algorithms, the seal never gets much better at balancing the pole than will be seen after about 900 trials." Geoffrey S. Knauth geoff@bos.camex.com Camex, Inc., 75 Kneeland St. geoff%bos.camex@uunet.uu.net Boston, MA 02111, (617) 426-3577 x451 --standard disclaimers--