home *** CD-ROM | disk | FTP | other *** search
- Path: sparky!uunet!olivea!spool.mu.edu!news.nd.edu!mentor.cc.purdue.edu!noose.ecn.purdue.edu!lips1.ecn.purdue.edu!kavuri
- From: kavuri@lips1.ecn.purdue.edu (Surya N Kavuri )
- Newsgroups: comp.ai.neural-nets
- Subject: A training problem
- Message-ID: <1993Jan12.005843.10910@noose.ecn.purdue.edu>
- Date: 12 Jan 93 00:58:43 GMT
- Sender: news@noose.ecn.purdue.edu (USENET news)
- Organization: Purdue University Engineering Computer Network
- Lines: 60
-
-
- Here I have a rpoblem with one input and 3 outputs.
- There are twenty training patterns.
-
- Input patterns are:
-
- 0.47500000000000
- 0.42500000000000
- 0.37500000000000
- 0.32500000000000
- 0.27500000000000
- 0.22500000000000
- 0.17500000000000
- 0.12500000000000
- 0.07500000000000
- 0.02500000000000
- -0.02500000000000
- -0.07500000000000
- -0.12500000000000
- -0.17500000000000
- -0.22500000000000
- -0.27500000000000
- -0.32500000000000
- -0.37500000000000
- -0.42500000000000
- -0.47500000000000
-
- Output patterns are:
-
- 5.22497730616910 -0.27581710755783 -0.47309647675852
- 4.65416529622998 -0.18871696832904 -0.32369758936110
- 4.08682235275143 -0.11129462234789 -0.19089857834116
- 3.52294847573346 -0.04355006961439 -0.07469944369871
- 2.96254366517605 0.01451668987146 0.02489981456624
- 2.40560792107921 0.06290565610968 0.10789919645370
- 1.85214124344295 0.10161682910025 0.17429870196367
- 1.30214363226725 0.13065020884318 0.22409833109614
- 0.75561508755213 0.15000579533847 0.25729808385113
- 0.21255560929758 0.15968358858611 0.27389796022862
- -0.32703480249641 0.15968358858611 0.27389796022862
- -0.86315614782982 0.15000579533847 0.25729808385113
- -1.39580842670266 0.13065020884318 0.22409833109614
- -1.92499163911493 0.10161682910025 0.17429870196367
- -2.45070578506663 0.06290565610968 0.10789919645370
- -2.97295086455776 0.01451668987146 0.02489981456624
- -3.49172687758832 -0.04355006961439 -0.07469944369871
- -4.00703382415831 -0.11129462234789 -0.19089857834116
- -4.51887170426773 -0.18871696832904 -0.32369758936110
- -5.02724051791657 -0.27581710755783 -0.47309647675852
-
- The data is so set up that input is orthogonal to outputs 2 and 3
- so that a linear method will not be able to fit this.
- Can you develop a network that will fit this ?
-
- How reliable are the weights connecting to these outputs ?
- (ie., if I use different subsets of inputs, can I expect
- small fluctuations ?)
-
-
- Surya Kavuri
-