home *** CD-ROM | disk | FTP | other *** search
- Path: sparky!uunet!dtix!darwin.sura.net!jvnc.net!gmd.de!jargon!al
- From: al@jargon.gmd.de (Alexander Linden)
- Newsgroups: comp.ai.neural-nets
- Subject: Re: Network Inversion
- Message-ID: <al.714491544@jargon>
- Date: 22 Aug 92 13:52:24 GMT
- References: <BtCqHo.AtH.1@cs.cmu.edu> <37125@sdcc12.ucsd.edu>
- Sender: news@gmd.de (USENET News)
- Organization: GMD, Sankt Augustin, Germany
- Lines: 352
- Nntp-Posting-Host: jargon
-
- demers@cs.ucsd.edu (David DeMers) writes:
-
- >In article <BtCqHo.AtH.1@cs.cmu.edu> tjochem+@CS.CMU.EDU (Todd Jochem) writes:
- >>I'm looking for references to network inversion papers. The basic idea
-
- The method Todd asked for is gradient descent in input
- space with respect to some error measure at the output.
-
- >There has been a fair amount of work done on this. Or at least
- >See Jordan, Michael I. & David Rumelhart, "Forward Models: Supervised
- >Learning with a Distal Teacher" (probably in Cognitive Science
-
- The work of Jordan and others David DeMers cites here IS quite related,
- but they do not work in input space rather they utilize the gradient
- in input space of an action model network (or transition model) as the
- error information at the output layer of a second net, the controller.
- This kind of procedure has also used by Nguyen/Widrow (1989) IJCNN, and
- Werbos. An earlier publication of very related stuff is
-
- @INPROCEEDINGS{Ersue87Y13,
- AUTHOR = {Ers\"u, E. and Tolle, H.},
- TITLE = {Hierarchical Learning Control -- An Appraoch with
- Neuron-like Associative Memories},
- BOOKTITLE = {Proceedings of IEEE Conference on Neural
- Information Processing, Denver, November 8-12},
- YEAR = 1987,
- EDITOR = {Anderson, D.},
- ORGANIZATION = {IEEE},
- PUBLISHER = {AIP},
- ADDRESS = {Denver, CO},
- KEYWORDS = {associative memory | hierarchical learning control},
- REF = {Y13}
- }
-
- >Mike Dyer and Risto Mikkulainen have also looked at the
- >input deltas for some purpose which I've forgotten, also
- >the cite, but should be easy to find...
-
- They have done work in the finding of proper coding schemes. Their work
- is published in
-
- @INPROCEEDINGS{Miikkulainen88Q52,
- AUTHOR = {Miikkulainen, R. and Dyer, M.~G.},
- TITLE = {Forming Global Representations with Extended
- Backpropagation},
- BOOKTITLE = {Proceedings ICNN-88, San Diego},
- YEAR = 1988,
- PAGES = {285-292},
- REF = {Q52}
- }
-
- and also in the connectionist summer school proceedings of 1989.
-
- >Mike Rossen in NIPS 3 has a paper on closed form inversion
-
- I did not see that this method calculates the gradient in input space.
- A direct calculation of the input seems not possible because it is
- often a many to one--mapping. Also the thresholds are not considered in
- his work...
-
- A small history of inversion of neural nets:
- --------------------------------------------
-
- The first publication I have found on this topic was Williams (1986).
- He trained a backprop net with the NETTALK database and inverted the
- mapping by backpropagating to the input layer.
-
- Independently to his very early work in the ancient times of
- connectionism other detected the potentials of inversion...
-
- Miikkulainen and Dyer (1988,1989) Forming Input Representations
- Kindermann and Linden (1988, 1989) Minimal Microfeature detection,
- experiments with recognition of handwritten digits.
-
- Thrun and Linden (1990) extended inversion to recurrent neural nets (by
- unfolding)
-
- Inversion of a forward model in neurocontrol, can be used to
- generate goal directed behaviour. It can determine the action
- which produces a successor state which is as close as possible to the
- systems goal (due to gradient descent the can be local minima)
-
- Thrun, Moeller and Linden (1990) derived lookahead planning by inversion of
- forward models and applied it to
- target tracking and cartpole ballancing.
- Kawato et al (1990) developed this idea for trajectory formation.
-
- Hwang et al (1990) used the idea of inversion to establish query
- learning. As pointed out in Linden/Kindermann (1989, IJCNN) it is easy
- possible to generate counterexamples by inversion. These
- counterexamples are very near at the class boundaries, thus with human
- interaction (telling the correct classification and extending the
- training set) the class boundaries can be moved.
-
- Hoskins et al (1992) applied this inversion to inverse kinematics.
- Also we did some experiments with this (Linden and Weber (1992)).
-
- Thrun and Moeller (1992) and Linden and Weber (1992) did experiments
- with exploration by utilizing gradient descent in input space, where
- a minimization of a competence model is used to guide some autonomous
- agent into regions which are believed unknown.
-
- Also Ron Chrisley (1990) has thought a bit on inversion in the context
- of neurocontrol. Also to mention is Steven Suddarth (1989).
-
- Here is a list of references (UNSORTED!!!):
-
- @INPROCEEDINGS{Williams86Z23,
- AUTHOR = {Williams, R. J.},
- TITLE = {Inverting a Connectionist Network Mapping by
- Backpropagation of Error},
- BOOKTITLE = {8th Annual Conference of the Cognitive Science Society},
- YEAR = 1986,
- PUBLISHER = {Lawrence Erlbaum},
- ADDRESS = {Hillsdale, NJ},
- KEYWORDS = {backpropagation | inversion | natural language},
- REF = {Z23}
- }
-
- @ARTICLE{Hoskins92PP22,
- AUTHOR = {Hoskins, D. A. and Hwang, J. N. and Vagners, J.},
- TITLE = {Iterative Inversion of Neural Networks and its
- Application to Adaptive Control},
- JOURNAL = {IEEE Transactions on Neural Networks},
- YEAR = {1992},
- VOLUME = {3},
- NUMBER = {2},
- PAGES = {292 - 301},
- KEYWORDS = {},
- REF = {PP22}
- }
-
- @UNPUBLISHED{Hwang90EE22,
- AUTHOR = {Hwang, J. N. and Choi, J. J. and Oh, S. and Marks, R. J.},
- TITLE = {Query Learning Based on Boundary Search and Gradient
- Computation of Trained Multilayer Perceptrons},
- NOTE = {to appear in Proceedings of IJCNN 90, San Diego, June 17-21,
- 1990},
- YEAR = {1990},
- KEYWORDS = {inversion},
- REF = {EE22}
- }
-
- @INPROCEEDINGS{Hwang90EE23,
- AUTHOR = {Hwang, J. N. and Chan, C. H.},
- TITLE = {Iterative Constrained Inversion of Neural Networks and its Applications},
- BOOKTITLE = {24th Conference on Information Systems and Sciences, Priceton,
- March 1990},
- YEAR = {1990},
- KEYWORDS = {inversion},
- REF = {EE23}
- }
-
- @INPROCEEDINGS{Miikkulainen88Q52,
- AUTHOR = {Miikkulainen, R. and Dyer, M.~G.},
- TITLE = {Forming Global Representations with Extended
- Backpropagation},
- BOOKTITLE = {Proceedings ICNN-88, San Diego},
- YEAR = 1988,
- PAGES = {285-292},
- REF = {Q52}
- }
-
-
- @INPROCEEDINGS{Linden88Correlative,
- AUTHOR = {Linden, A. and Kindermann, J.},
- TITLE = {A Correlative View on Backpropagation},
- BOOKTITLE = {Connectionism in Perspective},
- PAGES = {377-384},
- EDITOR = {Pfeifer, R. and Schreter, Z. and Fogelman, F. and Steels, L.},
- PUBLISHER = {Elsevier},
- ADDRESS = {Amsterdam},
- YEAR = 1989,
- NOTE = {Proceedings of the International Conference at the University
- of Z{\"u}rich, 10.-13.\ October 1988}
- }
-
-
- @INPROCEEDINGS{Kindermann89Microfeatures,
- AUTHOR = {Kindermann, J. and Linden, A.},
- TITLE = {Detection of Minimal Microfeatures by Internal Feedback},
- EDITOR = {Retti, J. and Leidlmair, K.},
- BOOKTITLE = {5. {\"O}sterreichische Artificial-Intelligence Tagung},
- YEAR = 1989,
- PUBLISHER = {Springer},
- ADDRESS = {Berlin, Heidelberg}
- }
-
- @INPROCEEDINGS{Linden89Inversion,
- AUTHOR = {Linden, A. and Kindermann, J.},
- TITLE = {Inversion of Multilayer Nets},
- BOOKTITLE = {Proceedings of the First International Joint Conference on
- Neural Networks, Washington DC},
- PUBLISHER = {IEEE},
- ADDRESS = {San Diego},
- YEAR = 1989,
- REF = {Inversion}
- }
-
-
- @MASTERSTHESIS{Linden90diplom,
- AUTHOR = {Linden, A.},
- TITLE = {{U}ntersuchung von {B}ackpropagation in konnektionistischen
- {S}ystemen},
- SCHOOL = {Universit{\"a}t Bonn, Informatik-Institutsbericht Nr. 80},
- TYPE = {Diplomarbeit},
- YEAR = {1990},
- REF = {diplom}
- }
-
- @ARTICLE{Kindermann90PC,
- AUTHOR = {Kindermann, J. and Linden, A.},
- TITLE = {Inversion of Neural Networks by Gradient Descent},
- JOURNAL = {Journal of Parallel Computing},
- YEAR = {1990},
- VOLUME = {14},
- NUMBER = {3},
- PAGES = {277-286},
- KEYWORDS = {},
- REF = {PC}
- }
-
- @INPROCEEDINGS{Thrun90Inversion,
- AUTHOR = {Thrun, S. and Linden, A.},
- TITLE = {Inversion in Time},
- BOOKTITLE = {Proceedings of the EURASIP Workshop on Neural Networks},
- ADDRESS = {Sesimbra, Portugal},
- YEAR = 1990,
- REF = {Inversion}
- }
-
- @INPROCEEDINGS{Thrun90Mental,
- AUTHOR = {Thrun, S. and Linden, A. and M{\"o}ller, K.},
- TITLE = {Planning with a mental modell},
- BOOKTITLE = {Proceedings of NeuroNimes 90, France},
- YEAR = {1990},
- REF = {Mental}
- }
-
- @INPROCEEDINGS{Thrun91NIPS,
- AUTHOR = {Thrun, S. and M{\"o}ller, K. and Linden, A.},
- TITLE = {Planning with an Adaptive World Model},
- BOOKTITLE = {Advances in Neural Information Processing Systems 3},
- YEAR = {1991},
- EDITOR = {Touretzky, D. S. and Lippmann, R.},
- PAGES = {},
- ORGANIZATION = {},
- PUBLISHER = {Morgan Kaufmann},
- ADDRESS = {San Mateo},
- MONTH = {},
- KEYWORDS = {},
- REF = {NIPS}
- }
-
- @INCOLLECTION{Kindermann92IEEE,
- AUTHOR = {Kindermann, J. and Linden, A.},
- TITLE = {Inversion of Neural Networks by Gradient Descent},
- BOOKTITLE = {Artificial Neural Networks: Concepts and Control
- Applications},
- EDITOR = {Vemuri, R.},
- PUBLISHER = {IEEE Computer Society Press},
- YEAR = {1992},
- REF = {Kindermann92IEEE}
- }
-
-
- @INPROCEEDINGS{Linden92AIREADER,
- AUTHOR = {Linden, A. and Weber, F.},
- TITLE = {Exploring the Kinematics of a Robot Arm by Competence
- Modelling },
- EDITOR = {Becker, B. and Gordon, T.},
- BOOKTITLE = {AI-Reader},
- PUBLISHER = {Erscheint als Arbeitspapier der GMD},
- YEAR = {1992}
- }
-
- @INPROCEEDINGS{Thrun92OO8,
- AUTHOR = {Thrun, S. and M\"oller, K.},
- TITLE = {Active Exploration in Dynamic Environments},
- BOOKTITLE = {Advances in Neural Information Processing Systems 4},
- YEAR = {1992},
- EDITOR = {Moody, J. E. and Hanson, S. J. and Lippmann, R. P.},
- PAGES = {},
- ORGANIZATION = {},
- PUBLISHER = {Morgan Kaufmann},
- ADDRESS = {},
- MONTH = {},
- KEYWORDS = {},
- REF = {OO8}
- }
-
- @UNPUBLISHED{Suddarth89Z4,
- AUTHOR = {Suddarth, S. C. and Bourrely, J. C.},
- TITLE = {A Back-Propagation Associative Memory for Both Positive and
- Negative Learning},
- YEAR = 1989,
- NOTE = {Poster Presentation at IJCNN-89},
- KEYWORDS = {backpropagation | associative memory | negative training},
- REF = {Z4}
- }
-
-
- @ARTICLE{Kawato90JJ16,
- AUTHOR = {Kawato, M. and Maeda, Y. and Uno, Y. and Suzuki, R.},
- TITLE = {Trajectory Formation of Arm Movement by Cascade
- Neural Network Model Based on Minimum Torque-Change
- Criterion},
- JOURNAL = {Biological Cybernetics},
- YEAR = {1990},
- VOLUME = {62},
- NUMBER = {},
- PAGES = {275 - 288},
- KEYWORDS = {},
- REF = {JJ16}
- }
-
-
- @UNPUBLISHED{Chrisley90JJ19,
- AUTHOR = {Chrisley, R.~L.},
- TITLE = {Cognitive Map Construction and Use: A Parallel
- Distributed Processing Approach},
- NOTE = {To appear in: Touretzky, D.~S. and Elman, J.~L.,
- Sejnowski, T.~J. and Hinton, G.~E. (eds.) (1990)
- Proceedings of the 1990 Connectionist Models Summer
- School. San Mateo, CA: Morgan Kaufmann.},
- YEAR = {1990},
- KEYWORDS = {},
- REF = {JJ19}
- }
-
-
-
- @INPROCEEDINGS{Rossen90NN31,
- AUTHOR = {Rossen, M. L.},
- TITLE = {Closed-Form Inversion of Backpropagation Networks: Theory
- and Optimiation Issues},
- BOOKTITLE = {Conference for Neural Information Processing (NIPS)},
- YEAR = {1990},
- EDITOR = {},
- PAGES = {},
- ORGANIZATION = {},
- PUBLISHER = {},
- ADDRESS = {Denver},
- MONTH = {November},
- KEYWORDS = {},
- REF = {NN31}
- }
- --
- Alexander Linden | TEL. (49 or 0) 2241/14-2435, FAX. -2618 or -2889
- GMD - AI Research Division | TELEX 889469 gmd d
- P. O. BOX 1316 | email: A.Linden@gmd.de
- D-5205 Sankt Augustin 1, FRG
-