1. B. DasGupta, H.T. Siegelmann, and E.D. Sontag. On the Intractability of Loading Neural Networks. In V. P. Roychowdhury, Siu K. Y., and Orlitsky A., editors, Theoretical Advances in Neural Computation and Learning, pages 357-389. Kluwer Academic Publishers, 1994. [PDF] Keyword(s): analog computing, neural networks, computational complexity, machine learning.



2. W. Maass, G. Schnitger, and E.D. Sontag. A comparison of the computational power of sigmoid and Boolean threshold circuits. In V. P. Roychowdhury, Siu K. Y., and Orlitsky A., editors, Theoretical Advances in Neural Computation and Learning, pages 127-151. Kluwer Academic Publishers, 1994. [PDF] Keyword(s): machine learning, neural networks, boolean systems. Abstract:

   We examine the power of constant depth circuits with sigmoid threshold gates for computing boolean functions. It is shown that, for depth 2, constant size circuits of this type are strictly more powerful than constant size boolean threshold circuits (i.e. circuits with linear threshold gates). On the other hand it turns out that, for any constant depth d, polynomial size sigmoid threshold circuits with polynomially bounded weights compute exactly the same boolean functions as the corresponding circuits with linear threshold gates.




3. F. Albertini and E.D. Sontag. Further results on controllability properties of discrete-time nonlinear systems. Dynam. Control, 4(3):235-253, 1994. [PDF] [doi:http://dx.doi.org/10.1007/BF01985073] Keyword(s): discrete-time, nonlinear control. Abstract:

   Controllability questions for discrete-time nonlinear systems are addressed in this paper. In particular, we continue the search for conditions under which the group-like notion of transitivity implies the stronger and semigroup-like property of forward accessibility. We show that this implication holds, pointwise, for states which have a weak Poisson stability property, and globally, if there exists a global "attractor" for the system.




4. F. Albertini and E.D. Sontag. State observability in recurrent neural networks. Systems Control Lett., 22(4):235-244, 1994. [PDF] [doi:http://dx.doi.org/10.1016/0167-6911(94)90054-X] Keyword(s): machine learning, neural networks, recurrent neural networks, observability, identifiability. Abstract:

   This paper concerns recurrent networks x'=s(Ax+Bu), y=Cx, where s is a sigmoid, in both discrete time and continuous time. Our main result is that observability can be characterized, if one assumes certain conditions on the nonlinearity and on the system, in a manner very analogous to that of the linear case. Recall that for the latter, observability is equivalent to the requirement that there not be any nontrivial A-invariant subspace included in the kernel of C. We show that the result generalizes in a natural manner, except that one now needs to restrict attention to certain special "coordinate" subspaces.




5. R. Koplon, E.D. Sontag, and M. L. J. Hautus. Observability of linear systems with saturated outputs. Linear Algebra Appl., 205/206:909-936, 1994. [PDF] Keyword(s): observability, saturation, bounded inputs. Abstract:

   In this paper, we present necessary and sufficient conditions for observability of the class of output-saturated systems. These are linear systems whose output passes through a saturation function before it can be measured.




6. H. T. Siegelmann and E.D. Sontag. Analog computation via neural networks. Theoret. Comput. Sci., 131(2):331-360, 1994. [PDF] [doi:http://dx.doi.org/10.1016/0304-3975(94)90178-3] Keyword(s): analog computing, neural networks, computational complexity, super-Turing computation, recurrent neural networks, neural networks, computational complexity. Abstract:

   We consider recurrent networks with real-valued weights. If allowed exponential time for computation, they turn out to have unbounded power. However, under polynomial-time constraints there are limits on their capabilities, though being more powerful than Turing Machines. Moreover, there is a precise correspondence between nets and standard non-uniform circuits with equivalent resources, and as a consequence one has lower bound constraints on what they can compute. We note that these networks are not likely to solve polynomially NP-hard problems, as the equality "P=NP" in our model implies the almost complete collapse of the standard polynomial hierarchy. We show that a large class of different networks and dynamical system models have no more computational power than this neural (first-order) model with real weights. The results suggest the following Church-like Thesis of Time-bounded Analog Computing: "Any reasonable analog computer will have no more power (up to polynomial time) than first-order recurrent networks."




7. H.J. Sussmann, E.D. Sontag, and Y. Yang. A general result on the stabilization of linear systems using bounded controls. IEEE Trans. Automat. Control, 39(12):2411-2425, 1994. [PDF] Keyword(s): saturation, neural networks, global stability, nonlinear stability, bounded inputs. Abstract:

   We present two constructions of controllers that globally stabilize linear systems subject to control saturation. We allow essentially arbitrary saturation functions. The only conditions imposed on the system are the obvious necessary ones, namely that no eigenvalues of the uncontrolled system have positive real part and that the standard stabilizability rank condition hold. One of the constructions is in terms of a "neural-network type" one-hidden layer architecture, while the other one is in terms of cascades of linear maps and saturations.

1. Y. Chitour, W. Liu, and E.D. Sontag. On the continuity and incremental gain properties of certain saturated linear feedback loops. In Proc. IEEE Conf. Decision and Control, Orlando, Dec. 1994, IEEE Publications, 1994, pages 127-132, 1994. [PDF] Keyword(s): saturation, bounded inputs.



2. B. DasGupta, H. T. Siegelmann, and E.D. Sontag. On a learnability question associated to neural networks with continuous activations (extended abstract). In COLT '94: Proceedings of the seventh annual conference on Computational learning theory, New York, NY, USA, pages 47-56, 1994. ACM Press. [doi:http://doi.acm.org/10.1145/180139.181009] Keyword(s): machine learning, analog computing, neural networks, computational complexity.



3. R. Koplon and E.D. Sontag. Techniques for parameter reconstruction in Fourier-Neural recurrent networks. In Proc. IEEE Conf. Decision and Control, Orlando, Dec. 1994, IEEE Publications, 1994, pages 213-218, 1994. Keyword(s): machine learning, neural networks, recurrent neural networks.



4. Y. Lin and E.D. Sontag. On control-Lyapunov functions under input constraints. In Proc. IEEE Conf. Decision and Control, Orlando, Dec. 1994, IEEE Publications, 1994, pages 640-645, 1994. Keyword(s): control-Lyapunov functions.



5. Y. Lin, E.D. Sontag, and Y. Wang. Recent results on Lyapunov-theoretic techniques for nonlinear stability. In Proc. Amer. Automatic Control Conf., Baltimore, June 1994, pages 1771-1775, 1994.



6. E.D. Sontag and Y. Wang. Notions equivalent to input-to-state stability. In Proc. IEEE Conf. Decision and Control, Orlando, Dec. 1994, IEEE Publications, 1994, pages 3438-3443, 1994. Keyword(s): input to state stability, ISS.



7. E.D. Sontag and Y. Wang. Orders of I/O equations and uniformly universal inputs. In Proc. IEEE Conf. Decision and Control, Orlando, Dec. 1994, IEEE Publications, 1994, pages 1270-1275, 1994. Keyword(s): identifiability, observability, realization theory.

This material is presented to ensure timely dissemination of scholarly and technical work. Copyright and all rights therein are retained by authors or by other copyright holders.

This document was translated from BibT_EX by bibtex2html