Abstract
Neural networks are usually considered as naturally parallel computing models. But the number of operators and the complex connection graph of standard neural models can not be handled by digital hardware devices. Though programmable digital hardware now stand as a real opportunity for flexible hardware implementations of neural networks, many area and topology problems arise when standard neural models are implemented onto programmable circuits such as FPGAs, so that the fast FPGA technology improvements can not be fully exploited. The theoretical and practical framework first introduced in [21] reconciles simple hardware topologies with complex neural architectures, thanks to some configurable hardware principles applied to neural computation: Field Programmable Neural Arrays (FPNA) lead to powerful neural architectures that are easy to map onto FPGAs, by means of a simplified topology and an original data exchange scheme. This two-chapter study gathers the different results that have been published about the FPNA concept, as well as some unpublished ones. This first part focuses on definitions and theoretical aspects. Starting from a general two-level definition of FPNAs, all proposed computation schemes are together described and compared. Their correctness and partial equivalence is justified. The computational power of FPNA-based neural networks is characterized through the concept of underparameterized convolutions.
Access this chapter
Tax calculation will be finalised at checkout
Purchases are for personal use only
Preview
Unable to display preview. Download preview PDF.
References
D. Abramson, K. Smith, P. Logothetis, and D. Duke. FPGA based implementation of a hopfield neural network for solving constraint satisfaction problems. In Proc. EuroMicro, 1998.
D. Anguita, S. Bencetti, A. De Gloria, G. Parodi, D. Ricci, and S. Ridella. FPGA implementation of high precision feedforward networks. In Proc. MicroNeuro, pages 240–243, 1997.
N. Avellana, A. Strey, R. Holgado, A. Fernandez, R. Capillas, and E. Valderrama. Design of a low-cost and high-speed neurocomputer system. In Proc. MicroNeuro, pages 221–226, 1996.
S.L. Bade and B.L. Hutchings. FPGA-based stochastic neural networks-implementation. In Proceedings of the IEEE Workshop on FPGAs for Custom Computing Machines, pages 189–198, 1994.
R. Baron and B. Girau. Parameterized normalization: application to wavelet networks. In Proc. IJCNN, volume 2, pages 1433–1437. IEEE, 1998.
J.-L. Beuchat. Conception d’un neuroprocesseur reconfigurable proposant des algorithmes d’apprentissage et d’elagage: une premiere etude. In Proc. NSI Neurosciences et Sciences de l’Ingenieur, 1998.
Y. Boniface. A parallel simulator to build distributed neural algorithms. In International Joint Conference on Neural Networks-IJCNN’01, Washington, USA, 2001.
N.M. Botros and M. Abdul-Aziz. Hardware implementation of an artificial neural network. In Proc. ICNN, volume 3, pages 1252–1257, 1993.
Y.K. Choi, K.H. Ahn, and S.-Y. Lee. Effects of multiplier output offsets on on-chip learning for analog neuro-chips. Neural Processing Letters, 4:1–8, 1996.
V.F. Cimpu. Hardware FPGA implementation of a neural network. In Proc. Int. Conf. Technical Informatics, volume 2, pages 57–68, 1996.
J.G. Eldredge and B.L. Hutchings. RRANN: a hardware implementation of the backpropagation algorithm using reconfigurable FPGAs. In Proceedings of the IEEE World Conference on Computational Intelligence, 1994.
A. Elisseeff and H. Paugam-Moisy. Size of multilayer networks for exact learning: analytic approach. Technical Report 96-16, LIP-ENSL, 1996.
W. Eppler, T. Fisher, H. Gemmeke, T. Becher, and G. Kock. High speed neural network chip on PCI-board. In Proc. MicroNeuro, pages 9–17, 1997.
S.K. Foo, P. Saratchandran, and N. Sundararajan. Parallel implementation of backpropagation neural networks on a heterogeneous array of transputers. IEEE Trans. on Systems, Man, and Cybernetics—Part B: Cybernetics, 27(1):118–126, 1997.
D. Franco and L. Carro. FPGA architecture comparison for nonconventional signal processing. In Proc. IJCNN, 2000.
K.-I. Funahashi. On the approximate realization of continuous mappings by neural networks. Neural Networks, 2:183–192, 1989.
R. Gadea, J. Cerda, F. Ballester, and A. Mocholi. Artificial neural network implementation on a single FPGA of a pipelined on-line backpropagation. In Proc. ISSS, pages 225–230, 2000.
C. Gegout, B. Girau, and F. Rossi. A general feedforward neural network model. Technical report NC-TR-95-041, NeuroCOLT, Royal Holloway, University of London, 1995.
C. Gegout, B. Girau, and F. Rossi. Generic back-propagation in arbitrary feedforward neural networks. In Artificial Neural Nets and Genetic Algorithms — Proc. of ICANNGA, pages 168–171. Springer-Verlag, 1995.
B. Girau. Dependencies of composite connections in Field Programmable Neural Arrays. Research report NC-TR-99-047, NeuroCOLT, Royal Holloway, University of London, 1999.
B. Girau. Du parallelisme des modeles connexionnistes a leur implantation parallele. PhD thesis n° 99ENSL0116, ENS Lyon, 1999.
B. Girau. Building a 2D-compatible multilayer neural network. In Proc. IJCNN. IEEE, 2000.
B. Girau. Conciliating connectionism and parallel digital hardware. Parallel and Distributed Computing Practices, special issue on Unconventional parallel architectures, 3(2):291–307, 2000.
B. Girau. Digital hardware implementation of 2D compatible neural networks. In Proc. IJCNN. IEEE, 2000.
B. Girau. FPNA: interaction between FPGA and neural computation. Int. Journal on Neural Systems, 10(3):243–259, 2000.
B. Girau. Neural networks on FPGAs: a survey. In Proc. Neural Computation, 2000.
B. Girau. Simplified neural architectures for symmetric boolean functions. In Proc. ESANN European Symposium on Artificial Neural Networks, pages 383–388, 2000.
B. Girau. On-chip learning of FPGA-inspired neural nets. In Proc. IJCNN. IEEE, 2001.
B. Girau and A. Tisserand. MLP computing and learning on FPGA using on-line arithmetic. Int. Journal on System Research and Information Science, special issue on Parallel and Distributed Systems for Neural Computing, 9(2–4), 2000.
C. Grassmann and J.K. Anlauf. Fast digital simulation of spiking neural networks and neuromorphic integration with spikelab. International Journal of Neural Systems, 9(5):473–478, 1999.
M. Gschwind, V. Salapura, and O. Maisch berger. A generic building block for Hopfield neural networks with on-chip learning. In Proc. ISCAS, 1996.
H. Hikawa. Frequency-based multilayer neural network with on-chip learning and enhanced neuron characteristics. IEEE Trans. on Neural Networks, 10(3):545–553, 1999.
R. Hoffmann, H.F. Restrepo, A. Perez-Uribe, C. Teuscher, and E. Sanchez. Implémentation d’un réseau de neurones sur un réseau de fpga. In Proc. Sympa’6, 2000.
P.W. Hollis, J.S. Harper, and J.J. Paulos. The effects of precision constraints in a backpropagation learning algorithm. Neural Computation, 2:363–373, 1990.
K. Hornik. Approximation capabilities of multilayer feedforward networks. Neural Networks, 4:251–257, 1991.
K. Hornik, M. Stinchcombe, and H. White. Multilayer feedforward networks are universal approximators. Neural Networks, 2:359–366, 1989.
N. Izeboudjen, A. Farah, S. Titri, and H. Boumeridja. Digital implementation of artificial neural networks: From VHDL description to FPGA implementation. In Proc. IWANN, 1999.
A. Johannet, L. Personnaz, G. Dreyfus, J.D. Gascuel, and M. Weinfeld. Specification and implementation of a digital Hopfield-type associative memory with on-chip training. IEEE Trans. on Neural Networks, 3, 1992.
J. Kennedy and J. Austin. A parallel architecture for binary neural networks. In Proc. MicroNeuro, pages 225–231, 1997.
K. Kollmann, K. Riemschneider, and H.C. Zeidler. On-chip backpropagation training using parallel stochastic bit streams. In Proc. MicroNeuro, pages 149–156, 1996.
A. Kramer. Array-based analog computation: principles, advantages and limitations. In Proc. MicroNeuro, pages 68–79, 1996.
V. Kumar, S. Shekhar, and M.B. Amin. A scalable parallel formulation of the back-propagation algorithm for hypercubes and related architectures. IEEE Transactions on Parallel and Distributed Systems, 5(10):1073–1090, October 1994.
P. Lysaght, J. Stockwood, J. Law, and D. Girma. Artificial neural network implementation on a fine-grained FPGA. In Proc. FPL, pages 421–432, 1994.
Y. Maeda and T. Tada. FPGA implementation of a pulse density neural network using simultaneous perturbation. In Proc. IJCNN, 2000.
S. McLoone and G.W. Irwin. Fast parallel off-line training of multilayer perceptrons. IEEE Trans. on Neural Networks, 8(3):646–653, 1997.
I. Milosavlevich, B. Flower, and M. Jabri. PANNE: a parallel computing engine for connectionist simulation. In Proc. MicroNeuro, pages 363–368, 1996.
P.D. Moerland and E. Fiesler. Hardware-friendly learning algorithms for neural networks: an overview. In Proc. MicroNeuro, 1996.
A. Montalvo, R. Gyurcsik, and J. Paulos. Towards a general-purpose analog VLSI neural network with on-chip learning. IEEE Trans. on Neural Networks, 8(2):413–423, 1997.
U.A. M-uller, A. Gunzinger, and W. Guggenb-uhl. Fast neural net simulation with a DSP processor array. IEEE Trans. on Neural Networks, 6(1):203–213, 1995.
T. Nordstrom and B. Svensson. Using and designing massively parallel computers for artificial neural networks. Journal of Parallel and Distributed Computing, 14(3):260–285, 1992.
R. Ostermark. A flexible multicomputer algorithm for artificial neural networks. Neural Networks, 9(1):169–178, 1996.
J. Park and I.W. Sandberg. Universal approximation using radial-basisfunction networks. Neural Computation, 3:246–257, 1991.
H. Paugam-Moisy. Optimal speedup conditions for a parallel back-propagation algorithm. In CONPAR, pages 719–724, 1992.
A. Perez-Uribe and E. Sanchez. FPGA implementation of an adaptablesize neural network. In Proc. ICANN. Springer-Verlag, 1996.
A. Petrowski. Choosing among several parallel implementations of the backpropagation algorithm. In Proc. ICNN, pages 1981–1986, 1994.
M. Rossmann, A. Buhlmeier, G. Manteuffel, and K. Goser. short-and long-term dynamics in a stochastic pulse stream neuron implemented in FPGA. In Proc. ICANN, LNCS, 1997.
M. Rossmann, T. Jost, A. Goser, K. B-uhlmeier, and G. Manteuffel. Exponential hebbian on-line lerarning implemented in FPGAs. In Proc. ICANN, 1996.
S. Sakaue, T. Kohda, H. Yamamoto, S. Maruno, and Shimeki Y. Reduction of required precision bits for back-propagation applied to pattern recognition. IEEE Trans. on Neural Networks, 4(2):270–275, 1993.
V. Salapura. Neural networks using bit-stream arithmetic: a space efficient implementation. In Proc. IEEE Int. Conf. on Circuits and Systems, 1994.
V. Salapura, M. Gschwind, and O. Maisch berger. A fast FPGA implementation of a general purpose neuron. In Proc. FPL, 1994.
K.M. Sammut and S.R. Jones. Arithmetic unit design for neural accelerators: cost performance issues. IEEE Trans. on Computers, 44(10), 1995.
M. Schaefer, T. Schoenauer, C. Wolff, G. Hartmann, H. Klar, and U. Ruckert. Simulation of spiking neural networks-architectures and implementations. Neurocomputing, (48):647–679, 2002.
S. Shams and J.-L. Gaudiot. Parallel implementations of neural networks. Int. J. on Artificial Intelligence, 2(4):557–581, 1993.
S. Shams and J.-L. Gaudiot. Implementing regularly structured neural networks on the DREAM machine. IEEE Trans. on Neural Networks, 6(2):407–421, 1995.
K. Siu, V. Roychowdhury, and T. Kailath. Depth-size tradeoffs for neural computation. IEEE Trans. on Computers, 40(12):1402–1412, 1991.
T. Szabo, L. Antoni, G. Horvath, and B. Feher. A full-parallel digital implementation for pre-trained NNs. In Proc. IJCNN, 2000.
M. van Daalen, P. Jeavons, and J. Shawe-Taylor. A stochastic neural architecture that exploits dynamically reconfigurable FPGAs. In Proc. of IEEE Workshop on FPGAs for Custom Computing Machines, pages 202–211, 1993.
M. Viredaz, C. Lehmann, F. Blayo, and P. Ienne. MANTRA: a multimodel neural network computer. In VLSI for Neural Networks and Artificial Intelligence, pages 93–102. Plenum Press, 1994.
J. Wawrzynek, K. Asanovic, and N. Morgan. The design of a neuromicroprocessor. IEEE Trans. on Neural Networks, 4(3):394–399, 1993.
Xilinx, editor. The Programmable Logic Data Book. Xilinx, 2002.
Q. Zhang and A. Benveniste. Wavelet networks. IEEE Trans. on Neural Networks, 3(6):889–898, Nov. 1992.
X. Zhang, M. McKenna, J.J. Mesirov, and D.L. Waltz. The backpropagation algorithm on grid and hypercube architectures. Parallel Computing, 14:317–327, 1990.
Author information
Authors and Affiliations
Editor information
Editors and Affiliations
Rights and permissions
Copyright information
© 2006 Springer
About this chapter
Cite this chapter
Girau, B. (2006). FPNA: Concepts and Properties. In: Omondi, A.R., Rajapakse, J.C. (eds) FPGA Implementations of Neural Networks. Springer, Boston, MA . https://doi.org/10.1007/0-387-28487-7_3
Download citation
DOI: https://doi.org/10.1007/0-387-28487-7_3
Publisher Name: Springer, Boston, MA
Print ISBN: 978-0-387-28485-9
Online ISBN: 978-0-387-28487-3
eBook Packages: EngineeringEngineering (R0)