Reservoir computing with dissociated neuronal culture
-
1
The University of Tokyo, Research Center for Advanced Science and Technology, Japan
Motivation
The Darwinian principle may operate in the cerebral cortex, in which the sources of intelligence are diversification and selection of neural activity patterns [1]. From diversified, chaotic activities of dissociated neuronal culture, we attempted to extract a coherent temporal pattern by utilizing an algorithm of reservoir computing, an emerging field of machine learning of recurrent neural network [2].
Material and Methods
A closed loop system with neuronal cultures was developed, as shown in Fig. 1, in order to implement the first-order reduced and controlled error learning, or FORCE learning. In FORCE learning, arbitrary temporal patterns can be generated from a recurrent neural network by linear combination of neural activities of neurons and providing feedback stimulation. The weights of the linear combination were optimized.
Cells from E18 rat cortices were dissociated in trypsin, and grown over MEA (Standard MEA 60MEA200/30iR-Ti-gr; Multichannel Systems) for 3 weeks or more. Neural signals were amplified with 1,100 gain, filtered with a passband of 1 ? 3 kHz, and recorded with a data acquisition device (NI CB-68LP; National instruments). Feedback stimulation was provided optically; laser illumination (473 nm; 500 mW) controlled by digital mirror device (DMD) activated RuBi-glutamate dissolved in the culture medium (100 ƒÊM) to release excitatory neurotransmitter. The magnitude of feedback stimulation was modulated by pulse duration.
Results
Figure 2 shows a typical example of FORCE learning of neuronal culture when the target was a constant value. At the early stage of learning, the output from the closed loop system fluctuated largely around the target value and the weight changed over time, while the fluctuation of both output and weight becomes small at the end of learning. In successful learning trials, when the weights were fixed constant as learned, the output stayed around the target. Thus, the present system was able to extract a coherent output from chaotic activities of neuronal cultures.
We then tested whether a problem-solving ability could emerge from this system. A mobile robot was controlled in a maze such that the robot moved straight forward if the extracted output matched the target, but turned left or right depending on the error between the output and target. In addition, disturbance stimuli were provided when the goal was out of the sight of robot, i.e., when the direction of goal was not placed within 90 degree in front of the robot, or when the robot detected obstacles with equipped sensors. With this simple strategy, we demonstrated that the robot were able to avoid obstacles and find the goal, supporting that FORCE learning underlies the emergent problem-solving ability in the brain.
Discussion
Some past studies operated a mobile robot on the basis of bidirectional communication with neuronal cultures, and such a neurorobotic Braitenberg vehicle exhibited an obstacle avoidance behavior [3] [4]. These studies demonstrated that an adequate sensori-motor coupling in a robot was successfully achieved through a Hebbian learning. The present study is completely different from these past studies in that neural plasticity is not assumed in the culture. Instead, the present study was inspired by homeostasis of neural networks, in which some aspects of neural activities, i.e., a constant value of firing rate in this experiment, are maintained by a closed loop system as far as the system is alive. Thus, we believe that both the Hebbian learning and homeostatic properties should be taken into account in the embodiment of neuronal systems.
Conclusion
FORCE learning, a kind of reservoir computing, was implemented with a primary dissociate culture serving as a reservoir. Our embodiment experiments demonstrated that a problem-solving ability emerges from this system.
References
[1] H. Takahashi, R. Yokota and R. Kanzaki (2013). Response variance in functional maps: neural darwinism revisited. Plos One 8: e68705
[2] D. Sussillo and L. F. Abbott (2009). Generating coherent patterns of activity from chaotic neural networks. Neuron 63: 544-557
[3] A. Novellino, P. D'Angelo, L. Cozzi, M. Chiappalone, V. Sanguineti and S. Martinoia (2007). Connecting neurons to a mobile robot: an in vitro bidirectional neural interface. Comput Intell Neurosci: Art no 12725
[4] D. J. Bakkum, Z. C. Chao and S. M. Potter (2008). Spatio-temporal electrical stimuli shape behavior of an embodied cortical network in a goal-directed learning task. J Neural Eng 5: 310-323.
Acknowledgements
This study was partially supported by Ashahi Glass Foundation, Kayamori Foundation and KAKENHI (26630089).
Keywords:
embodiment,
Optical stimulation,
primary culture,
Caged compound,
MEA
Conference:
MEA Meeting 2016 |
10th International Meeting on Substrate-Integrated Electrode Arrays, Reutlingen, Germany, 28 Jun - 1 Jul, 2016.
Presentation Type:
Poster Presentation
Topic:
MEA Meeting 2016
Citation:
Takahashi
H,
Yasuda
S,
Yada
Y and
Kanzaki
R
(2016). Reservoir computing with dissociated neuronal culture.
Front. Neurosci.
Conference Abstract:
MEA Meeting 2016 |
10th International Meeting on Substrate-Integrated Electrode Arrays.
doi: 10.3389/conf.fnins.2016.93.00027
Copyright:
The abstracts in this collection have not been subject to any Frontiers peer review or checks, and are not endorsed by Frontiers.
They are made available through the Frontiers publishing platform as a service to conference organizers and presenters.
The copyright in the individual abstracts is owned by the author of each abstract or his/her employer unless otherwise stated.
Each abstract, as well as the collection of abstracts, are published under a Creative Commons CC-BY 4.0 (attribution) licence (https://creativecommons.org/licenses/by/4.0/) and may thus be reproduced, translated, adapted and be the subject of derivative works provided the authors and Frontiers are attributed.
For Frontiers’ terms and conditions please see https://www.frontiersin.org/legal/terms-and-conditions.
Received:
22 Jun 2016;
Published Online:
24 Jun 2016.
*
Correspondence:
Dr. Hirokazu Takahashi, The University of Tokyo, Research Center for Advanced Science and Technology, Tokyo, Japan, takahashi@i.u-tokyo.ac.jp