Abstract
We analyze the computational power of discrete-time recurrent neural networks (NNs) with the saturated-linear activation function within the Chomsky hierarchy. This model restricted to integer weights coincides with binary-state NNs with the Heaviside activation function, which are equivalent to finite automata (Chomsky level 3), while rational weights make this model Turing complete even for three analog-state units (Chomsky level 0). For an intermediate model \(\alpha \)ANN of a binary-state NN that is extended with \(\alpha \ge 0\) extra analog-state neurons with rational weights, we have established the analog neuron hierarchy 0ANNs \(\subset \) 1ANNs \(\subset \) 2ANNs \(\subseteq \) 3ANNs. The separation 1ANNs \(\subsetneqq \) 2ANNs has been witnessed by the deterministic context-free language (DCFL) \(L_\#=\{0^n1^n\,|\,n\ge 1\}\) which cannot be recognized by any 1ANN even with real weights, while any DCFL (Chomsky level 2) is accepted by a 2ANN with rational weights. In this paper, we generalize this result by showing that any non-regular DCFL cannot be recognized by 1ANNs with real weights, which means (DCFLs \(\setminus \) REG\()\,\subset \,(\)2ANNs \(\setminus \) 1ANNs), implying 0ANNs \(=\) 1ANNs \(\cap \) DCFLs. For this purpose, we show that \(L_\#\) is the simplest non-regular DCFL by reducing \(L_\#\) to any language in this class, which is by itself an interesting achievement in computability theory.
Research was done with institutional support RVO: 67985807 and partially supported by the grant of the Czech Science Foundation GA19-05704S.
Access this chapter
Tax calculation will be finalised at checkout
Purchases are for personal use only
Notes
- 1.
- 2.
In online input/output protocols, the time between reading two consecutive input symbols as well as the delay in outputting the result after an input has been read, is bounded by a constant, while in offline protocols these time intervals are not bounded.
References
Alon, N., Dewdney, A.K., Ott, T.J.: Efficient simulation of finite automata by neural nets. J. ACM 38(2), 495–514 (1991)
Balcázar, J.L., Gavaldà, R., Siegelmann, H.T.: Computational power of neural networks: A characterization in terms of Kolmogorov complexity. IEEE Trans. Inf. Theory 43(4), 1175–1183 (1997)
Horne, B.G., Hush, D.R.: Bounds on the complexity of recurrent neural network implementations of finite state machines. Neural Netw. 9(2), 243–252 (1996)
Indyk, P.: Optimal simulation of automata by neural nets. In: Mayr, E.W., Puech, C. (eds.) STACS 1995. LNCS, vol. 900, pp. 337–348. Springer, Heidelberg (1995). https://doi.org/10.1007/3-540-59042-0_85
Jančar, P., Mráz, F., Plátek, M., Vogel, J.: Restarting automata. In: Reichel, H. (ed.) FCT 1995. LNCS, vol. 965, pp. 283–292. Springer, Heidelberg (1995). https://doi.org/10.1007/3-540-60249-6_60
Jančar, P., Mráz, F., Plátek, M., Vogel, J.: On monotonic automata with a restart operation. J. Automata Lang. Comb. 4(4), 287–311 (1999)
Kilian, J., Siegelmann, H.T.: The dynamic universality of sigmoidal neural networks. Inf. Comput. 128(1), 48–56 (1996)
Koiran, P.: A family of universal recurrent networks. Theoret. Comput. Sci. 168(2), 473–480 (1996)
Minsky, M.: Computations: Finite and Infinite Machines. Prentice-Hall, Englewood Cliffs (1967)
Orponen, P.: Computing with truly asynchronous threshold logic networks. Theoret. Comput. Sci. 174(1–2), 123–136 (1997)
Schmidhuber, J.: Deep learning in neural networks: An overview. Neural Netw. 61, 85–117 (2015)
Siegelmann, H.T.: Recurrent neural networks and finite automata. J. Comput. Intell. 12(4), 567–574 (1996)
Siegelmann, H.T.: Neural Networks and Analog Computation: Beyond the Turing Limit. Birkhäuser, Boston (1999). https://doi.org/10.1007/978-1-4612-0707-8
Siegelmann, H.T., Sontag, E.D.: Analog computation via neural networks. Theoret. Comput. Sci. 131(2), 331–360 (1994)
Siegelmann, H.T., Sontag, E.D.: On the computational power of neural nets. J. Comput. Syst. Sci. 50(1), 132–150 (1995)
Šíma, J.: Analog stable simulation of discrete neural networks. Neural Netw. World 7(6), 679–686 (1997)
Šíma, J.: Energy complexity of recurrent neural networks. Neural Comput. 26(5), 953–973 (2014)
Šíma, J.: Three analog neurons are Turing universal. In: Fagan, D., Martín-Vide, C., O’Neill, M., Vega-Rodríguez, M.A. (eds.) TPNC 2018. LNCS, vol. 11324, pp. 460–472. Springer, Cham (2018). https://doi.org/10.1007/978-3-030-04070-3_36
Šíma, J.: Counting with analog neurons. In: Tetko, I.V., Kůrková, V., Karpov, P., Theis, F. (eds.) ICANN 2019. LNCS, vol. 11727, pp. 389–400. Springer, Cham (2019). https://doi.org/10.1007/978-3-030-30487-4_31
Šíma, J.: Subrecursive neural networks. Neural Netw. 116, 208–223 (2019)
Šíma, J., Orponen, P.: General-purpose computation with neural networks: a survey of complexity theoretic results. Neural Comput. 15(12), 2727–2778 (2003)
Šíma, J., Savický, P.: Quasi-periodic \(\beta \)-expansions and cut languages. Theoret. Comput. Sci. 720, 1–23 (2018)
Šíma, J., Wiedermann, J.: Theory of neuromata. J. ACM 45(1), 155–178 (1998)
Šorel, M., Šíma, J.: Robust RBF finite automata. Neurocomputing 62, 93–110 (2004)
Author information
Authors and Affiliations
Corresponding author
Editor information
Editors and Affiliations
Rights and permissions
Copyright information
© 2019 Springer Nature Switzerland AG
About this paper
Cite this paper
Šíma, J., Plátek, M. (2019). One Analog Neuron Cannot Recognize Deterministic Context-Free Languages. In: Gedeon, T., Wong, K., Lee, M. (eds) Neural Information Processing. ICONIP 2019. Lecture Notes in Computer Science(), vol 11955. Springer, Cham. https://doi.org/10.1007/978-3-030-36718-3_7
Download citation
DOI: https://doi.org/10.1007/978-3-030-36718-3_7
Published:
Publisher Name: Springer, Cham
Print ISBN: 978-3-030-36717-6
Online ISBN: 978-3-030-36718-3
eBook Packages: Computer ScienceComputer Science (R0)