Abstract
We proclaim the use of application-specific instruction set processors with programmable accuracy called Anytime Instruction Processors (AIPs) for Convolutional Neural Network (CNN) inference. For a floating-point operation, the number of correctly computed mantissa result bits can be freely adjusted, allowing for a fine-grained trade-off analysis between accuracy, execution time and energy. We propose a Design Space Exploration (DSE) technique in which the accuracy of CNN computations is determined layer-by-layer. As one result, we show that reductions of up to 62% in energy consumption are achievable for a representative ResNet-18 benchmark in comparison to a solution where each layer is computed at full accuracy according to the IEEE 754 single precision floating-point format.
Access this chapter
Tax calculation will be finalised at checkout
Purchases are for personal use only
Notes
- 1.
Note that \(|V|=21\) for ResNet-18 when also counting three 1 \(\times \) 1 convolutional layers.
References
Alizadeh, M., Behboodi, A., van Baalen, M., Louizos, C., Blankevoort, T., Welling, M.: Gradient \(\ell _{1}\) regularization for quantization robustness. In: 8th International Conference on Learning Representations, ICLR (2020)
Brand, M., Witterauf, M., Bosio, A., Teich, J.: Anytime floating-point addition and multiplication-concepts and implementations. In: 31st IEEE International Conference on Application-Specific Systems, Architectures and Processors, ASAP, pp. 157–164. IEEE (2020)
Brand, M., Witterauf, M., Hannig, F., Teich, J.: Anytime instructions for programmable accuracy floating-point arithmetic. In: Proceedings of the 16th ACM International Conference on Computing Frontiers, CF, pp. 215–219. ACM (2019)
Cai, Y., Yao, Z., Dong, Z., Gholami, A., Mahoney, M.W., Keutzer, K.: ZeroQ: a novel zero shot quantization framework. In: 2020 IEEE/CVF Conference on Computer Vision and Pattern Recognition, CVPR 2020, pp. 13166–13175 (2020)
Cheng, Y., Wang, D., Zhou, P., Zhang, T.: Model compression and acceleration for deep neural networks: the principles, progress, and challenges. IEEE Sig. Process. Mag. 35(1), 126–136 (2018)
Deb, K., Agrawal, S., Pratap, A., Meyarivan, T.: A fast and elitist multiobjective genetic algorithm: NSGA-II. IEEE Trans. Evol. Comput. 6(2), 182–197 (2002)
Fu, Y., et al.: FracTrain: fractionally squeezing bit savings both temporally and spatially for efficient DNN training. In: Advances in Neural Information Processing Systems 33: Annual Conference on Neural Information Processing Systems 2020, NeurIPS, pp. 12127–12139 (2020)
Garófalo, A., Rusci, M., Conti, F., Rossi, D., Benini, L.: PULP-NN: accelerating quantized neural networks on parallel ultra-low-power RISC-V processors. Phil. Trans. R. Soc. A 378, 20190155 (2019)
Genc, H., et al.: Gemmini: an agile systolic array generator enabling systematic evaluations of deep-learning architectures. CoRR abs/1911.09925 (2019)
He, K., Zhang, X., Ren, S., Sun, J.: Deep residual learning for image recognition. In: 2016 IEEE Conference on Computer Vision and Pattern Recognition, CVPR, pp. 770–778. IEEE Computer Society (2016)
Heidorn, C., Witterauf, M., Hannig, F., Teich, J.: Efficient mapping of CNNs onto tightly coupled processor arrays. J. Comput. 14(8), 541–556 (2019)
Keszocze, O., Brand, M., Witterauf, M., Heidorn, C., Teich, J.: Aarith: an arbitrary precision number library. In: ACM/SIGAPP Symposium On Applied Computing (2021). Aarith is publicly available at https://github.com/keszocze/aarith
Krizhevsky, A., Nair, V., Hinton, G.: The CIFAR-10 dataset (2010). https://www.cs.toronto.edu/~kriz/cifar.html. Accessed 20 Feb 2020
LeCun, Y., Cortes, C., Burges, C.J.: MNIST handwritten digit database (1998). http://yann.lecun.com/exdb/mnist/. Accessed 18 Nov 2020
Lukasiewycz, M., Glaß, M., Reimann, F., Teich, J.: Opt4j: a modular framework for meta-heuristic optimization. In: Proceedings of the 13th Annual Conference on Genetic and Evolutionary Computation, pp. 1723–1730 (2011)
Mrazek, V., Vasícek, Z., Sekanina, L., Hanif, M.A., Shafique, M.: ALWANN: automatic layer-wise approximation of deep neural network accelerators without retraining. In: Proceedings of the International Conference on Computer-Aided Design, ICCAD 2019, pp. 1–8. ACM (2019)
la Parra, C.D., Guntoro, A., Kumar, A.: ProxSim: GPU-based simulation framework for cross-layer approximate DNN optimization. In: 2020 Design, Automation & Test in Europe Conference & Exhibition, DATE 2020, pp. 1193–1198. IEEE (2020)
Rek, P., Sekanina, L.: TypeCNN: CNN development framework with flexible data types. In: Design, Automation & Test in Europe Conference & Exhibition, DATE 2019, pp. 292–295 (2019)
Synopsys: Synopsys 32/28 nm and 90 nm generic libraries (2020). https://www.synopsys.com/community/university-program/teaching-resources.html. Accessed 06 Aug 2020
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., Rasnayake, L., Själander, M.: BISMO: a scalable bit-serial matrix multiplication overlay for reconfigurable computing. In: 28th International Conference on Field Programmable Logic and Applications, FPL 2018, pp. 307–314. IEEE Computer Society (2018)
Wang, K., Liu, Z., Lin, Y., Lin, J., Han, S.: HAQ: hardware-aware automated quantization with mixed precision. In: IEEE Conference on Computer Vision and Pattern Recognition, CVPR 2019, pp. 8612–8620 (2019)
Acknowledgments
This work was partially funded by the Deutsche Forschungsgemeinschaft (DFG, German Research Foundation)—Project Number 146371743—TRR 89 Invasive Computing and the German Federal Ministry for Education and Research (BMBF) within project KISS (01IS19070B).
Author information
Authors and Affiliations
Corresponding author
Editor information
Editors and Affiliations
Rights and permissions
Copyright information
© 2021 Springer Nature Switzerland AG
About this paper
Cite this paper
Schuster, A., Heidorn, C., Brand, M., Keszocze, O., Teich, J. (2021). Design Space Exploration of Time, Energy, and Error Rate Trade-offs for CNNs Using Accuracy-Programmable Instruction Set Processors. In: Kamp, M., et al. Machine Learning and Principles and Practice of Knowledge Discovery in Databases. ECML PKDD 2021. Communications in Computer and Information Science, vol 1524. Springer, Cham. https://doi.org/10.1007/978-3-030-93736-2_29
Download citation
DOI: https://doi.org/10.1007/978-3-030-93736-2_29
Published:
Publisher Name: Springer, Cham
Print ISBN: 978-3-030-93735-5
Online ISBN: 978-3-030-93736-2
eBook Packages: Computer ScienceComputer Science (R0)