Abstract
Learning compositional dynamics with recurrent neural networks (RNNs) trained with back-propagation through time (BPTT) is usually a difficult task. Typically RNNs learn the consecutive shape along target sequences from time step to time step, focusing on local temporal correlations. When the challenge is to identify and model independent, unknown data subcomponents, that is, data generating causes on-the-fly during training, however, this local temporal shape-oriented inductive learning bias is obstructive. We propose a modular, compositional RNN architecture and derive simple procedures to automatically infer the source subdynamics that generate the data. We show that the involved error signal separation can be used for both teacher forcing and model-distinct target signal provision in the compositional RNN architecture. As a result, the entire network is able to learn compositional dynamics, developing emergent, flexibly adaptable signal decompositions within the distributed modules. We demonstrate that in this way simple RNNs trained with BPTT can learn sequences that could so far only be solved effectively with reservoir computing approaches. Moreover we show that these RNNs are much more robust against signal noise when compared to traditional BPTT or reservoir computing approaches.
This is a preview of subscription content, log in via an institution.
Buying options
Tax calculation will be finalised at checkout
Purchases are for personal use only
Learn about institutional subscriptionsReferences
Butz, M.V., Kutter, E.F.: How the Mind Comes Into Being: Introducing Cognitive Science from a Functional and Computational Perspective. Oxford University Press, Oxford (2017)
Gao, Z., et al.: A cortico-cerebellar loop for motor planning. Nature 563(7729), 113 (2018)
He, K., Zhang, X., Ren, S., Sun, J.: Deep residual learning for image recognition. In: The IEEE Conference on Computer Vision and Pattern Recognition (CVPR), June 2016
Hochreiter, S., Schmidhuber, J.: Long short-term memory. Neural Comput. 9(8), 1735–1780 (1997). https://doi.org/10.1162/neco.1997.9.8.1735
Jaeger, H.: The “echo state” approach to analysing and training recurrent neural networks. Technical report, GMD Report, 148, Fraunhofer Institute for Analysis and Information Systems AIS, Sankt Augustin, Germany (2001)
Kandel, E.R., Schwartz, J.H., Jessell, T.M., Jessell, M.B.T., Siegelbaum, S., Hudspeth, A.: Principles of Neural Science. McGraw-hill, New York (2000)
Kingma, D.P., Ba, J.L.: Adam: a method for stochastic optimization. 3rd International Conference for Learning Representations abs/1412.6980 (2015)
Koryakin, D., Lohmann, J., Butz, M.V.: Balanced echo state networks. Neural Netw. 36, 35–45 (2012)
Otte, S., Krechel, D., Liwicki, M.: JANNLab neural network framework for Java. In: Poster Proceedings of MLDM 2013, pp. 39–46. ibai-publishing, New York (2013)
Otte, S., Butz, M.V., Koryakin, D., Becker, F., Liwicki, M., Zell, A.: Optimizing recurrent reservoirs with neuro-evolution. Neurocomputing 192, 128–138 (2016)
Pathak, J., Hunt, B., Girvan, M., Lu, Z., Ott, E.: Model-free prediction of large spatiotemporally chaotic systems from data: a reservoir computing approach. Phys. Rev. Lett. 120, 024102 (2018)
Schmidhuber, J., Wierstra, D., Gagliolo, M., Gomez, F.: Training recurrent neural networks by evolino. Neural Comput. 19, 757–779 (2007)
Svoboda, K., Li, N.: Neural mechanisms of movement planning: motor cortex and beyond. Curr. Opin. Neurobiol. 49, 33–41 (2018)
Werbos, P.: Backpropagation through time: what it does and how to do it. Proc. IEEE 78(10), 1550–1560 (1990). https://doi.org/10.1109/5.58337
Wu, Y., et al.: Google’s Neural Machine Translation System: Bridging the Gap between Human and Machine Translation. ArXiv e-prints 1609.08144 (2016)
Acknowledgments
The authors would like to thank Sander Bothé, CWI Amsterdam, for helpful comments and suggestions regarding this work.
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
Otte, S., Rubisch, P., Butz, M.V. (2019). Gradient-Based Learning of Compositional Dynamics with Modular RNNs. In: Tetko, I., Kůrková, V., Karpov, P., Theis, F. (eds) Artificial Neural Networks and Machine Learning – ICANN 2019: Theoretical Neural Computation. ICANN 2019. Lecture Notes in Computer Science(), vol 11727. Springer, Cham. https://doi.org/10.1007/978-3-030-30487-4_38
Download citation
DOI: https://doi.org/10.1007/978-3-030-30487-4_38
Published:
Publisher Name: Springer, Cham
Print ISBN: 978-3-030-30486-7
Online ISBN: 978-3-030-30487-4
eBook Packages: Computer ScienceComputer Science (R0)