Abstract
Techniques from differential topology are used to give polynomial bounds for the VC-dimension of sigmoidal neural networks. The bounds are quadratic in ω, the dimension of the space of weights. Similar results are obtained for a wide class of Pfaffian activation functions. The obstruction (in differential topology) to improving the bound to an optimal bound \({\cal O}(w log w)\)(ω log ω) is discussed, and attention is paid to the role of other parameters involved in the network architecture.
Research partially supported by the DFG Grant KA 673/4-1, and by ESPRIT BR Grants 7097 and ECUS 030.
Research supported in part by a Senior Research Fellowship of the SERC.
Preview
Unable to display preview. Download preview PDF.
References
M. Anthony, N. Biggs, Computational Learning Theory: An Introduction, Cambridge University Press, 1992.
M. Anthony, J. Shawe-Taylor, A Result of Vapnik with Applications, Discrete Applied Math. 47 (1993), pp. 207–217.
A. Borodin, P. Tiwari, On the Decidability of Sparse Univariate Polynomial Interpolation, Proc. 22nd ACM STOC (1990), pp. 535–545.
L. van den Dries, Tame Topology and 0-minimal Structures, preprint, University of Illinois, Urbana, 1992; to appear as a book.
L. van den Dries, A.Macintyre and D.Marker, The Elementary Theory of Restricted Analytic Fields with Exponentation, Annuals of Mathematics 140 (1994), pp 183–205.
P.Goldberg and M.Jerrum, Bounding the Vapnik Chervonenkis Dimension of Concept Classes Parametrized by Real Numbers. Machine Learning, 1994 (to appear). A preliminary version appeared in Proc. 6th ACM Workshop on Computational Learning Theory, pp. 361–369, 1993.
G.H. Hardy, Properties of Logarithmic-Exponential Functions, Proc. London Math. Soc. 10 (1912), pp. 54–90.
D. Haussler, Decision Theoretic Generalizations of the PAC Model for Neural Nets and other Learning Applications, Information an Computation 100, (1992), pp. 78–150.
J. Hertz, A. Krogh and R. G. Palmer, Introduction to the Theory of Neural Computation, Addison-Wesley, 1991.
M. W. Hirsch, Differential Topology, Springer-Verlag, 1976.
M. Karpinski and A. Macintyre, Quadratic Bounds for VC Dimension at Sigmoidal Neural Networks, Research Report No. 85116-CS, Universität Bonn, 1994; to be submitted.
M.Karpinski and T.Werther, VC Dimension and Uniform Learnability of Sparse Polynomials and Rational Functions, SIAM J. Computing 22 (1993), pp 1276–1285.
A.G.Khovanski, Fewnomials, American Mathematical Society, Providence, R.I., 1991.
J.Knight, A.Pillay and C.Steinhorn, Definable Sets and Ordered Structures II, Trans. American Mathematical Society 295 (1986), pp.593–605.
M.C.Laskowsky, Vapnik-Chervonenkis Classes od Definable Sets, J.London Math. Society 45 (1992), pp 377–384.
W.Maass, On the Complexity of Learning on Feedforward Neural Nets, in Proc. EATCS Advanced School on Computational Learning and Cryptography, Vietri sul Mare, 1993.
W. Maass, G. Schnitger and E. D. Sontag, On the Computational Power of Sigmoidal versus Boolean Threshold Circuits, Proc. 32nd IEEE FOGS (1991), pp. 767–776.
A.J.Macintyre and E.D.Sontag, Finiteness results for Sigmoidal Neural Networks, Proc. 25th ACM STOC (1993), pp.325–334.
J.Milnor, On the Betti Numbers of Real Varieties, Proc. of the American Mathematical Society 15 (1964), pp 275–280.
J.Milnor, Topology from the Differentiable Viewpoint, Univ.Press, Virginia, 1965.
J. Shawe-Taylor, Sample Sizes for Sigmoidal Neural Networks, Preprint, University of London, 1994.
G. Turan and F. Vatan, On the Computation of Boolean Functions by Analog Circuits of Bounded Fan-in, Proc. 35th IEEE FOCS (1994), pp. 553–564.
H.E.Warren, Lower Bounds for Approximation by Non-linear Manifolds, Trans. of the AMS 133 (1968), pp. 167–178.
A.J.Wilkie, Model Completeness Results of Restricted Pfaffian Functions and the Exponential Function, to appear in Journal of the AMS, 1994.
Author information
Authors and Affiliations
Editor information
Rights and permissions
Copyright information
© 1995 Springer-Verlag Berlin Heidelberg
About this paper
Cite this paper
Karpinski, M., Macintyre, A. (1995). Bounding VC-dimension for neural networks: Progress and prospects. In: Vitányi, P. (eds) Computational Learning Theory. EuroCOLT 1995. Lecture Notes in Computer Science, vol 904. Springer, Berlin, Heidelberg. https://doi.org/10.1007/3-540-59119-2_189
Download citation
DOI: https://doi.org/10.1007/3-540-59119-2_189
Published:
Publisher Name: Springer, Berlin, Heidelberg
Print ISBN: 978-3-540-59119-1
Online ISBN: 978-3-540-49195-8
eBook Packages: Springer Book Archive