×

zbMATH — the first resource for mathematics

A model for learning to segment temporal sequences, utilizing a mixture of RNN experts together with adaptive variance. (English) Zbl 1254.68215
Summary: This paper proposes a novel learning method for a mixture of recurrent neural network (RNN) experts model, which can acquire the ability to generate desired sequences by dynamically switching between experts. Our method is based on maximum likelihood estimation, using a gradient descent algorithm. This approach is similar to that used in conventional methods; however, we modify the likelihood function by adding a mechanism to alter the variance for each expert. The proposed method is demonstrated to successfully learn Markov chain switching among a set of 9 Lissajous curves, for which the conventional method fails. The learning performance, analyzed in terms of the generalization capability, of the proposed method is also shown to be superior to that of the conventional method. With the addition of a gating network, the proposed method is successfully applied to the learning of sensory-motor flows for a small humanoid robot as a realistic problem of time series prediction and generation.

MSC:
68T05 Learning and adaptive systems in artificial intelligence
65C40 Numerical analysis or methods applied to Markov chains
Software:
LSTM
PDF BibTeX XML Cite
Full Text: DOI
References:
[1] Bengio, Y.; Simard, P.; Frasconi, P., Learning long-term dependencies with gradient descent is difficult, IEEE transactions on neural networks, 5, 2, 157-166, (1994)
[2] Elman, J.L., Finding structure in time, Cognitive science, 14, 179-211, (1990)
[3] Evans, G., Semantic theory and tacit knowledge, (), 118-137
[4] Hammer, B.; Tin˘o, P., Recurrent neural networks with small weights implement definite memory machines, Neural computation, 15, 1897-1929, (2003) · Zbl 1085.68125
[5] Hichreiter, S.; Schmidhuber, J., Long short-term memory, Neural computation, 9, 8, 1735-1780, (1997)
[6] Igari, I., & Tani, J. (2008). Incremental learning of sequence patterns with a modular network model. Neurocomputing (in press)
[7] Ikeda, K.; Otsuka, K.; Matsumoto, K., Maxwell Bloch turbulence, Progress of theoretical physics (suppl.), 99, 295-324, (1989)
[8] Jacobs, R.A.; Jordan, M.I.; Nowlan, S.J.; Hinton, G.E., Adaptive mixtures of local experts, Neural computation, 3, 1, 79-87, (1991)
[9] Jaeger, H. (2001). Short term memory in echo state networks. GMD report, 152, pp. 1-60
[10] Jaeger, H.; Haas, H., Harnessing nonlinearity: predicting chaotic systems and saving energy in wireless communication, Science, 78-80, (2004)
[11] Jordan, M.I., Attractor dynamics and parallelism in a connectionist sequential machine, (), 531-546
[12] Jordan, M.I., Indeterminate motor skill learning problems, ()
[13] Jordan, M.I.; Jacobs, R.A., Hierarchical mixtures of experts and the EM algorithm, Neural computation, 6, 2, 181-214, (1994)
[14] Kaneko, K.; Tsuda, I., Chaotic itinerancy, Chaos: focus issue on chaotic itinerancy, 13, 926-936, (2003) · Zbl 1080.37531
[15] Kataoka, N.; Kaneko, K., Functional dynamics I: articulation process, Physica D, 138, 225-250, (2000) · Zbl 0955.37056
[16] Kataoka, N.; Kaneko, K., Functional dynamics II: syntactic structure, Physica D, 149, 174-196, (2001) · Zbl 0980.37011
[17] Kataoka, N.; Kaneko, K., Dynamical networks in function dynamics, Physica D, 181, 235-251, (2003) · Zbl 1025.37025
[18] Maass, W., Natschläger, T., & Markram, H. (2002). A fresh look at real-time computation in generic recurrent neural circuits. Technical report. Institute for Theoretical Computer Science, TU Graz
[19] Namikawa, J.; Hashimoto, T., Dynamics and computation in functional shifts, Nonlinearity, 17, 1317-1336, (2004) · Zbl 1051.68098
[20] Namikawa, J., & Tani, J. Learning to imitate stochastic time series in a compositional way by chaos. Neural Networks (submitted for publication)
[21] Pollack, J., The induction of dynamical recognizers, Machine learning, 7, 227-252, (1991)
[22] Rumelhart, D.E.; Hinton, G.E.; Williams, R.J., Learning internal representations by error propagation, (), 318-362
[23] Sato, Y.; Ikegami, T., Nonlinear computation with switching map systems, Journal of universal computer science, 6, 881-905, (2000) · Zbl 0971.68052
[24] Schmidhuber, J.; Gers, F.; Eck, D., Learning nonregular languages: A comparison of simple recurrent networks and LSTM, Neural computation, 14, 2039-2041, (2002) · Zbl 1010.68857
[25] Tani, J.; Nishimoto, R.; Namikawa, J.; Ito, M., Co-developmental learning between human and humanoid robot using a dynamic neural network model, IEEE transactions on system, man and cybernetics, 38, 1, 43-59, (2008)
[26] Tani, J.; Nolfi, S., Learning to perceive the world as articulated: an approach for hierarchical learning in sensory – motor systems, Neural networks, 12, 1131-1141, (1999)
[27] Williams, R.J.; Zipser, D., A learning algorithm for continually running fully recurrent neural networks, Neural computation, 1, 270-280, (1989)
[28] Wolpert, D.; Kawato, M., Multiple paired forward and inverse models for motor control, Neural networks, 11, 1317-1329, (1998)
This reference list is based on information provided by the publisher or from digital mathematics libraries. Its items are heuristically matched to zbMATH identifiers and may contain data conversion errors. It attempts to reflect the references listed in the original paper as accurately as possible without claiming the completeness or perfect precision of the matching.