Abstract
The application of Deep Neural Networks (DNN) for reinforcement learning has proven effective in solving complex problems, such as playing video games or training robots to perform human tasks. Training based on reinforcement implies the continuous interaction of the agent powered by the DNN and the environment, vanishing the typical separation between the training and inference stages in deep learning. However, the high memory and accuracy requirements of gradient-based training algorithms prevent using FPGAs for these applications. As an alternative, this work demonstrates the feasibility of using Evolutionary Algorithms (EA) for training DNNs and their usage in reinforcement learning scenarios. Unlike backpropagation, EA-based training of neural networks, referred to as neuroevolution, can be effectively implemented on FPGAs. Moreover, this paper shows how the inherent parallelism of EAs can be effectively exploited in multi-FPGA scenarios to accelerate the learning process. The proposed FPGA-based neuroevolutionary framework has been validated by building a system capable of learning autonomously to play the Pong Atari game in less than 25 generations.
This project has been funded by the Spanish Ministry for Science and Innovation under the project TALENT (ref. PID2020-116417RB-C42).
This is a preview of subscription content, log in via an institution to check access.
Access this chapter
Tax calculation will be finalised at checkout
Purchases are for personal use only
Similar content being viewed by others
References
Asseman, A., Antoine, N., Ozcan, A.S.: Accelerating deep neuroevolution on distributed fpgas for reinforcement learning problems. J. Emerg. Technol. Comput. Syst. 17(2) (2021). https://doi.org/10.1145/3425500
Brockman, G., et al.: Openai gym (2016)
Chen, T., et al.: Tvm: end-to-end optimization stack for deep learning. arXiv preprint arXiv:1802.04799, vol. 11, no. 20 (2018)
Digilent: Pynq-z1 board reference manual (2017). https://reference.digilentinc.com/_media/reference/programmable-logic/pynq-z1/pynq-rm.pdf. Accessed 8 July 2021
García, A., Zamacola, R., Otero, A., de la Torre, E.: A dynamically reconfigurable bbnn architecture for scalable neuroevolution in hardware. Electronics 9(5) (2020). https://doi.org/10.3390/electronics9050803, https://www.mdpi.com/2079-9292/9/5/803
Irmen: Pyro4 framework (2021). https://github.com/irmen/Pyro4. Accessed 8 July 2021
Kachris, C., Falsafi, B., Soudris, D.: Hardware Accelerators in Data Centers. Springer, Heidelberg (2019)
Koutník, J., Schmidhuber, J., Gomez, F.: Evolving deep unsupervised convolutional networks for vision-based reinforcement learning. In: Proceedings of the 2014 Annual Conference on Genetic and Evolutionary Computation, pp. 541–548. GECCO’14, Association for Computing Machinery, New York, NY, USA (2014). https://doi.org/10.1145/2576768.2598358
Liu, X., Xu, H., Liao, W., Yu, W.: Reinforcement learning for cyber-physical systems. In: 2019 IEEE International Conference on Industrial Internet (ICII), pp. 318–327. IEEE (2019)
Luo, C., Sit, M.K., Fan, H., Liu, S., Luk, W., Guo, C.: Towards efficient deep neural network training by fpga-based batch-level parallelism. J. Semicond. 41(2), 022403 (2020)
Mnih, V., et al.: Playing atari with deep reinforcement learning (2013)
Moreau, T., et al.: A hardware-software blueprint for flexible deep learning specialization. IEEE Micro 39(5), 8–16 (2019)
Nurvitadhi, E., et al.: Can fpgas beat gpus in accelerating next-generation deep neural networks? In: Proceedings of the 2017 ACM/SIGDA International Symposium on Field-programmable Gate Arrays, pp. 5–14 (2017)
Pappalardo, A.: Xilinx/brevitas (2021). https://doi.org/10.5281/zenodo.3333552
Parisi, G.I., Kemker, R., Part, J.L., Kanan, C., Wermter, S.: Continual lifelong learning with neural networks: A review. Neural Netw. 113, 54–71 (2019)
Petroski Such, F., Madhavan, V., Conti, E., Lehman, J., Stanley, K.O., Clune, J.: Deep neuroevolution: genetic algorithms are a competitive alternative for training deep neural networks for reinforcement learning. arXiv e-prints. arXiv:1712.06567 (Dec 2017)
Russell, S., Norvig, P.: Artificial intelligence: a modern approach (2002)
Sze, V., Chen, Y.H., Yang, T.J., Emer, J.S.: Efficient processing of deep neural networks: a tutorial and survey. Proc. IEEE 105(12), 2295–2329 (2017)
Umuroglu, Y., et al.: Finn: a framework for fast, scalable binarized neural network inference. In: Proceedings of the 2017 ACM/SIGDA international symposium on field-programmable gate arrays. pp. 65–74 (2017)
Author information
Authors and Affiliations
Corresponding author
Editor information
Editors and Affiliations
Rights and permissions
Copyright information
© 2022 The Author(s), under exclusive license to Springer Nature Switzerland AG
About this paper
Cite this paper
Laserna, J., Otero, A., Torre, E.d.l. (2022). A Multi-FPGA Scalable Framework for Deep Reinforcement Learning Through Neuroevolution. In: Gan, L., Wang, Y., Xue, W., Chau, T. (eds) Applied Reconfigurable Computing. Architectures, Tools, and Applications. ARC 2022. Lecture Notes in Computer Science, vol 13569. Springer, Cham. https://doi.org/10.1007/978-3-031-19983-7_4
Download citation
DOI: https://doi.org/10.1007/978-3-031-19983-7_4
Published:
Publisher Name: Springer, Cham
Print ISBN: 978-3-031-19982-0
Online ISBN: 978-3-031-19983-7
eBook Packages: Computer ScienceComputer Science (R0)