Papers
Topics
Authors
Recent
Gemini 2.5 Flash
Gemini 2.5 Flash
143 tokens/sec
GPT-4o
7 tokens/sec
Gemini 2.5 Pro Pro
46 tokens/sec
o3 Pro
4 tokens/sec
GPT-4.1 Pro
38 tokens/sec
DeepSeek R1 via Azure Pro
28 tokens/sec
2000 character limit reached

Adaptive control of recurrent neural networks using conceptors (2405.07236v1)

Published 12 May 2024 in cs.LG and nlin.AO

Abstract: Recurrent Neural Networks excel at predicting and generating complex high-dimensional temporal patterns. Due to their inherent nonlinear dynamics and memory, they can learn unbounded temporal dependencies from data. In a Machine Learning setting, the network's parameters are adapted during a training phase to match the requirements of a given task/problem increasing its computational capabilities. After the training, the network parameters are kept fixed to exploit the learned computations. The static parameters thereby render the network unadaptive to changing conditions, such as external or internal perturbation. In this manuscript, we demonstrate how keeping parts of the network adaptive even after the training enhances its functionality and robustness. Here, we utilize the conceptor framework and conceptualize an adaptive control loop analyzing the network's behavior continuously and adjusting its time-varying internal representation to follow a desired target. We demonstrate how the added adaptivity of the network supports the computational functionality in three distinct tasks: interpolation of temporal patterns, stabilization against partial network degradation, and robustness against input distortion. Our results highlight the potential of adaptive networks in machine learning beyond training, enabling them to not only learn complex patterns but also dynamically adjust to changing environments, ultimately broadening their applicability.

Definition Search Book Streamline Icon: https://streamlinehq.com
References (35)
  1. R. Berner, T. Gross, C. Kuehn, J. Kurths,  and S. Yanchuk, “Adaptive dynamical networks,” Physics Reports 1031, 1–59 (2023).
  2. T. Gross, C. J. D. D’Lima,  and B. Blasius, “Epidemic dynamics on an adaptive network,” Physical Review Letters 96, 208701 (2006).
  3. R. Berner, S. Yanchuk,  and E. Schöll, “What adaptive neuronal networks teach us about power grids,” Physical Review E 103, 042315 (2021).
  4. C. Clopath, L. Büsing, E. Vasilaki,  and W. Gerstner, “Connectivity reflects coding: a model of voltage-based stdp with homeostasis,” Nature Neuroscience 13, 344–352 (2010).
  5. G. B. Morales, C. R. Mirasso,  and M. C. Soriano, “Unveiling the role of plasticity rules in reservoir computing,” Neurocomputing 461, 705–715 (2021).
  6. Q. Xuan, F. Du, H. Dong, L. Yu,  and G. Chen, “Structural control of reaction-diffusion networks,” Physical Review E 84, 036101 (2011).
  7. H. Hewamalage, C. Bergmeir,  and K. Bandara, “Recurrent neural networks for time series forecasting: Current status and future directions,” International Journal of Forecasting 37, 388–427 (2021).
  8. I. Sutskever, J. Martens,  and G. E. Hinton, “Generating text with recurrent neural networks,” in Proceedings of the 28th international conference on machine learning (ICML-11) (2011) pp. 1017–1024.
  9. M. Ganaie, S. Ghosh, N. Mendola, M. Tanveer,  and S. Jalan, “Identification of chimera using machine learning,” Chaos: An Interdisciplinary Journal of Nonlinear Science 30 (2020).
  10. Y. Pan and J. Wang, “Model predictive control of unknown nonlinear dynamical systems based on recurrent neural networks,” IEEE Transactions on Industrial Electronics 59, 3089–3101 (2011).
  11. C. Hens, U. Harush, S. Haber, R. Cohen,  and B. Barzel, “Spatiotemporal signal propagation in complex networks,” Nature Physics 15, 403–412 (2019).
  12. H. Jaeger and H. Haas, “Harnessing nonlinearity: Predicting chaotic systems and saving energy in wireless communication,” Science 304, 78–80 (2004).
  13. M. Lukoševičius and H. Jaeger, “Reservoir computing approaches to recurrent neural network training,” Computer Science Review 3, 127–149 (2009).
  14. L. Appeltant, M. C. Soriano, G. Van Der Sande, J. Danckaert, S. Massar, J. Dambre, B. Schrauwen, C. R. Mirasso,  and I. Fischer, “Information processing using a single dynamical node as complex system,” Nat. Commun. 2, 468 (2011).
  15. D. Brunner, M. C. Soriano, C. R. Mirasso,  and I. Fischer, “Parallel photonic information processing at gigabyte per second data rates using transient states,” Nat. Commun 4, 1364 (2013).
  16. P.J. Werbos, “Backpropagation Through Time: What It Does and How to Do It,”  (1990).
  17. H. Jaeger, “The "echo state" approach to analysing and training recurrent neural networks,” GMD Report , 1–47 (2001).
  18. C. Klos, Y. F. K. Kossio, S. Goedeke, A. Gilra,  and R.-M. Memmesheimer, “Dynamical learning of dynamics,” Phys. Rev. Lett. 125, 088103 (2020).
  19. L.-W. Kong, H.-W. Fan, C. Grebogi,  and Y.-C. Lai, “Machine learning prediction of critical transition and system collapse,” Phys. Rev. Res. 3, 13090 (2021), 2012.01545 .
  20. J. Z. Kim, Z. Lu, E. Nozari, G. J. Pappas,  and D. S. Bassett, “Teaching recurrent neural networks to infer global temporal structure from local examples,” Nat. Mach. Intell. 3, 316–323 (2021).
  21. M. Goldmann, C. R. Mirasso, I. Fischer,  and M. C. Soriano, “Learn one size to infer all: Exploiting translational symmetries in delay-dynamical and spatiotemporal systems using scalable neural networks,” Phys. Rev. E 106, 044211 (2022).
  22. X. A. Ji and G. Orosz, “Learn from one and predict all: single trajectory learning for time delay systems,” Nonlinear Dynamics  (2024), 10.1007/s11071-023-09206-y.
  23. H. Jaeger, “Controlling recurrent neural networks by conceptors,” arXiv preprint arXiv:1403.3369  (2014).
  24. C. Fernando and S. Sojakka, “Pattern recognition in a bucket,” in European conference on artificial life (Springer, 2003) pp. 588–597.
  25. H. Jaeger, “Using Conceptors to Manage Neural Long-Term Memories for Temporal Patterns,” Journal of Machine Learning Research 18, 1–43 (2017).
  26. X. He and H. Jaeger, “Overcoming Catastrophic Interference using Conceptor-Aided Backpropagation,” in International Conference on Learning Representations (2018).
  27. J. Yperman and T. Becker, “Bayesian optimization of hyper-parameters in reservoir computing,” arXiv preprint arXiv:1611.05193  (2016).
  28. F. Wyffels, J. Li, T. Waegeman, B. Schrauwen,  and H. Jaeger, “Frequency modulation of large oscillatory neural networks,” Biological cybernetics 108, 145–157 (2014).
  29. A. Feragen and A. Fuster, “Geometries and Interpolations for Symmetric Positive Definite Matrices,” in Modeling, Analysis, and Visualization of Anisotropy, edited by T. Schultz, E. Özarslan,  and I. Hotz (Springer International Publishing, Cham, 2017) pp. 85–113, series Title: Mathematics and Visualization.
  30. C. Torres-Huitzil and B. Girau, “Fault and Error Tolerance in Neural Networks: A Review,” IEEE Access 5, 17322–17341 (2017).
  31. Carnegie Mellon University Graphics Lab, “Motion capture database,” Accessed: 2023-04-03.
  32. W. R. Ashby, Design for a Brain: The Origin of Adaptive Behavior (Martino Fine Books, 2014).
  33. I. Fischer, Y. Liu,  and P. Davis, “Synchronization of chaotic semiconductor laser dynamics on subnanosecond time scales and its potential for chaos communication,” Physical Review A 62, 011801(R) (2000).
  34. I. Kohler, “The formation and transformation of the perceptual world,” Psychological Issues 3, 1–173 (1963), place: US Publisher: International Universities Press, Inc.
  35. E. Di Paolo, “Homeostatic adaptation to inversion of the visual field and other sensorimotor disruptions,”   (2000).

Summary

We haven't generated a summary for this paper yet.

Ai Sparkles Streamline Icon: https://streamlinehq.com Summarize Now