Thursday , March 28 2024

Distributed Neural Networks Microcontroller Implementation and Applications

Ioan Susnea
University “Dunărea de Jos” of Galaţi, Romania
111, Domnească Street, Galaţi, 800201, ROMANIA

Abstract:

In this paper it is argued that, for any three-layer perceptron, it is always possible to design an equivalent distributed ANN, wherein the neurons are implemented on the nodes of a communication network, and the synapses between them are established in the communication process. In this approach, neurons are seen as processing and communication entities. Since both local and distributed implementations of a specific ANN are perfectly equivalent, they can use the same set of synapse weights, i.e. a distributed ANN can be trained on a local, equivalent software implementation. Two use cases are presented to demonstrate the validity of the idea.

Keywords:

Distributed ANN, microcontrollers, robot navigation, smart environment.

>>Full text
CITE THIS PAPER AS:
Ioan SUSNEA, Distributed Neural Networks Microcontroller Implementation and Applications, Studies in Informatics and Control, ISSN 1220-1766, vol. 21 (2), pp. 165-172, 2012. https://doi.org/10.24846/v21i2y201206

1. Introduction

Although artificial neural networks (ANNs) have been successfully used in almost any research field ([1], [2]), there are relatively few reports about microcontroller based implementations thereof ([3]).

While it is obvious that the severe limitations in what concerns hardware resources and computing power of the usual low cost microcontrollers make it difficult to use them for the implementation of ANNs, it is also clear that, in principle, it is possible to design systems wherein the computational task is divided in sub-tasks, executed by a plurality of microcontrollers, connected in a communication network ([4]).

Typical communication networks comprising nodes with limited data processing capabilities are the Wireless Sensor Networks (WSN – [5]). Several researchers noticed the similarities between WSN and ANNs, and proposed solutions for implementing neural network structures over WSN ([6], [7], [8]).

Another interesting research direction deriving from the concept of distributed neural network is the attempt to identify neural models of the interactions between agents in swarms ([9], [10).

In this paper, we go beyond the exploration of similarities and analogies between WSN and ANNs, and argue that, for any three-layer perceptron, it is always possible to design an equivalent distributed ANN. The neurons are implemented on the nodes of the communication network, and the synapses between them are defined in the process of communication.

Two use cases are presented to demonstrate the validity of the idea. In the first example, a mobile robot communicates with a plurality of “neural beacons”, or “neural landmarks”, and the resulting distributed ANN directly controls the navigation of the robot. In another example, a number of “smart” PIR (Passive Infrared) motion detectors, learn to control the HVAC system according to the activity patterns of the inhabitants, in order to reduce the overall energy requirements for heating the building.

Beyond this introduction, this paper is structured as follows:

Section II presents the principles of the implementation of distributed ANNs starting from the model of the three-layer perceptron.

Section III describes an example of using a distributed ANN, consisting in a number of “neural beacons” deployed in the environment, to control a mobile robot for path tracking.

Section IV presents a similar network, wherein the neurons are implemented by smart PIR sensors, used to adjust the reference temperature of a HVAC System in accordance with the occupancy patterns of the inhabitants.

Section V is reserved for conclusions.

References:

  1. WIDROW, B., D. E. RUMELHART MICHAEL, A. LEHR, Neural Networks: Applications in Industry, Business and Science, Communications of the ACM, Volume 37 Issue 3, March 1994.
  2. HAGAN, M., H. DEMUTH, Neural Networks for Control, Invited Tutorial, 1999 American Control Conference, San Diego, June, 1999, pp. 1642-1656.
  3. BINFET, J., B. M. WILAMOWSKI, Microprocessor Implementation of Fuzzy Systems and Neural Networks, in Neural Networks, 2001. Proceedings. IJCNN ’01. International Joint Conference on, July 2001, pp. 234-239.
  4. SOARES, S. G. et. al., Building Distributed Soft Sensors, International Journal of Computer Information Systems and Industrial Management Applications ISSN 2150-7988 vol. 3, 2011, pp. 202-209.
  5. VIEIRA, M. A. M., C. N. COELHO, Jr., D. C. da SILVA, Jr., J. M. da MATA, Survey on Wireless Sensor Network Devices, Emerging Technologies and Factory Automation, 2003. Proceedings. ETFA ’03. IEEE Conference Sept. 2003 pp. 537-544.
  6. OLDEWURTEL, F., P. MAHONEN, Neural Wireless Sensor Networks, Systems and Networks Communications, 2006. ICSNC ’06. International Conference on, pp.28-36.
  7. ENAMI, N., R. A. MOGHADAM, K. DADASHTABAR, Neural Network Based Energy Efficiency in Wireless Sensor Networks: A Survey, International Journal of Computer Science & Engineering Survey (IJCSES) Vol.1, No.1, August 2010.
  8. SHAH, K., M. KUMAR, S. INC, T. ADDISON, Resource Management in Wireless Sensor Networks using Collective Intelligence, International Conference on Intelligent Sensors, Sensor Networks and Information Processing, 2008. ISSNIP 2008, pp. 423-428.
  9. PARUNAK, H. v. D., S. BRUECKNER, The Cognitive Aptitude of Swarming Agents, Vector Research Center, Ann Arbor, MI, 2009. www. newvectors.net/staff/parunakv/CASA.pdf
  10. SUSNEA, I., G. VASILIU, A. FILIPESCU A. RADASCHIN, Virtual Pheromones for Real-Time Control of Autonomous Mobile Robots, in Studies of Informatics and Control, Vol 18, issue 3, 2009, pp. 233-240
  11. RUMELHART D. E., G. E. HINTON, R. J. WILLIAMS, Learning Representations by Back-propagating Errors, Nature 323, October 1986, pp. 533-536.
  12. BOSCH, R., CAN Specification Version 2.0,, Robert Bosch GmbH, Stuttgart, 1991.
  13. K.-R. SOHN, H.-J. LEE, Y. KIM, Wireless CAN Communications based on White LED, Third International Conference on Ubiquitous and Future Networks (ICUFN), 15-17 June 2011, pp. 127-130.
  14. BONASTRE, A., R. ORS, J.V. CAPELLA, J. HERRERO, Distribution of Neural-based Discrete Control Algorithms Applied to Home Automation with CAN, Proceedings of the 8th International CAN Conference, Las Vegas, USA, February 2002.
  15. CAPELLA, J. V., A. BONASTRE, R. ORS, An Advanced and Distributed Control Architecture Based on Intelligent Agents and Neural Networks, IEEE International Workshop on Intelligent Data Acquisition and Advanced Computing Systems: Technology and Applications 8-10 September 2003, Lvov, Ukraine.
  16. www.mobilerobots.com.
  17. MIGLINO, O., H. LUND, S. NOLFI, Evolving Mobile Robot in Simulated and Real Environment, Artificial Life, vol. 2, no. 4, 1995, pp. 417-434.
  18. HELLSTRÖM, T., O. RINGDAHL, Follow the Past: A Path-tracking Algorithm for Autonomous Vehicles, International Journal of Vehicle Autonomous Systems, Vol. 4, Nos. 2-4, 2006, pp.216-224.
  19. Directive 2010/31/Eu of the European Parliament and of the Council of 19 May 2010 on the Energy Performance Of Buildings, Official Journal of The European Union L153, Vol. 53, 18 June, 2010
  20. US patent US2008291036A1
  21. US patent US2009/266904A1