ABSTRACT
Stochastic circuits (SCs) offer significant area, power and energy benefits at the cost of computational inaccuracies. SCs have received particular attention recently in the context of neural networks (NNs). Many NNs use the maximum function, e.g., in the max-pooling layer of convolutional NNs. Currently, approximate workarounds are often employed for this function. We propose NMax, a new SC design for the maximum function that produces an exact result with latency similar to an approximate circuit. Furthermore, unlike most stochastic functions, NMax is correlation insensitive. We also observe that maximum calculations are subject to application-specific bias and analyze this bias.
- Alaghi A. and J.P. Hayes. Exploiting correlation in stochastic circuit design. Proc. Intl. Conf. on Computer Design (ICCD), 39--46, 2013.Google ScholarCross Ref
- Alaghi A., C. Li and J.P. Hayes. Stochastic circuits for real-time image-processing applications. Proc. Design Automation Conf. (DAC), art. 136, 2013. Google ScholarDigital Library
- Alaghi A. and J.P. Hayes. STRAUSS: Spectral transfrom use in stochastic circuit synthesis. IEEE Trans. Computer-Aided Design, 34.11: 1770--1783, 2015.Google ScholarCross Ref
- Alaghi A. and J.P. Hayes. Survey of stochastic computing. ACM Trans. Embedded Comp. Syst., 12: art. 92, 2013. Google ScholarDigital Library
- Box G.E.P. and G. Jenkins. Time Series Analysis: Forecasting and Control, Holden-Day, 1976. Google ScholarDigital Library
- Brown B.D. and H.C. Card. Stochastic neural computation I: Computational elements. IEEE Trans. Computers, 50: 891--905, 2001. Google ScholarDigital Library
- Canals V., A. Morro, A. Oliver, M.L. Alomar and J.L. Rosselló. A new stochastic computing methodology for efficient neural network implementation. IEEE Trans. Neural Networks and Learning Systems 27.3: 551--564, 2016.Google ScholarCross Ref
- Chen T. and J.P. Hayes. Analyzing and controlling accuracy in stochastic circuits. Proc. Intl. Conf. on Computer Design (ICCD), 367--373, 2014.Google ScholarCross Ref
- Farmahini-Farahani A., H.J. Duwe, Michael J. Schulte and K. Compton. Modular design of high-throughput, low-latency sorting units. IEEE Trans. Computers, 62, 1389--1402, 2013. Google ScholarDigital Library
- Gaines B.R. Stochastic computing systems. Advances in Information Systems Science, 2: 37--172, 1969.Google ScholarCross Ref
- Gupta P.K. and R. Kumaresan, Binary multiplication with PN sequences, IEEE Trans. Acoustics, Speech and Signal Processing, 36, 603--606, 1988.Google ScholarCross Ref
- Ichihara H., T. Sugino, S. Ishii, T. Iwagaki and T. Inoue. Compact and accurate digital filters based on stochastic computing. IEEE Trans. Emerging Topics in Computing, 7, 31--43, 2019.Google ScholarCross Ref
- Li P. and D.J. Lilja. Using stochastic computing to implement digital image processing algorithms. Proc. Intl. Conf. on Computer Design (ICCD),: 154--161, 2011. Google ScholarDigital Library
- Naderi A., S. Mannor, M. Sawan and W.J. Gross. Delayed stochastic decoding of LDPC codes. IEEE Trans. Signal Processing, 59: 5617--5626, 2011. Google ScholarDigital Library
- Najafi M., D.J. Lilja, M.D. Riedel and K. Bazargan. Low-cost sorting network circuits using unary processing. IEEE Trans. VLSI Systems, 26, 1471--1480, 2018.Google ScholarCross Ref
- Neugebauer F., I. Polian and J.P. Hayes. S-Box-based random number generation for stochastic computing. Proc. Conf. on Digital System Design (DSD), 1--6, 2017.Google Scholar
- Neugebauer F., I. Polian and J.P. Hayes. Framework for quantifying and managing accuracy in stochastic circuit design. Proc. DATE, 1--6, 2017. Google ScholarDigital Library
- Qian, W., X. Li, M.D. Riedel, K. Bazargan and D.J. Lilja. An architecture for fault-tolerant computation with stochastic logic. IEEE Trans. Computers 60, 93--105, 2011. Google ScholarDigital Library
- Ren A., Z. Li, C. Ding, Q. Qiu, Y. Wang, J. Li, X. Qian and B. Yuan. SC-DCNN: Highly scalable deep convolutional neural network using stochastic computing. Proc. ASPLOS, 405--418, 2017. Google ScholarDigital Library
- Stanford CS class, cs231n: Convolutional neural networks for visual recognition, 2016. http://cs231n.github.io/convolutional-networks/Google Scholar
- Ting P-S and J.P. Hayes. Isolation-based decorrelation of stochastic circuits. Proc. Intl. Conf. on Computer Design (ICCD), 88--95, 2016.Google ScholarCross Ref
- van Daalen, M., P. Jeavons, J.Shawe-Taylor and Dave Cohen. Device for generating binary sequences for stochastic computing. Electronics Letters, 29: 80--81, 1993.Google ScholarCross Ref
- Wang R., J. Han, B.F. Cockburn and D.G. Elliot. Design, evaluation and fault-tolerance analysis of stochastic FIR filters. Microelectronics Reliability 57: 111--127, 2016.Google ScholarCross Ref
- Yu J., K. Kim, J. Lee and K. Choi. Accurate and efficient stochastic computing hardware for convolutional neural networks. Proc. Intl. Conf. on Computer Design (ICCD), 104--112, 2017.Google ScholarCross Ref
Index Terms
- On the maximum function in stochastic computing
Recommendations
Neural Network Classifiers Using a Hardware-Based Approximate Activation Function with a Hybrid Stochastic Multiplier
Special Issue on Emerging Networks-on-Chip and Regular PapersNeural networks are becoming prevalent in many areas, such as pattern recognition and medical diagnosis. Stochastic computing is one potential solution for neural networks implemented in low-power back-end devices such as solar-powered devices and ...
A hardware implementation of a radial basis function neural network using stochastic logic
DATE '15: Proceedings of the 2015 Design, Automation & Test in Europe Conference & ExhibitionHardware implementations of artificial neural networks typically require significant amounts of hardware resources. This paper proposes a novel radial basis function artificial neural network using stochastic computing elements, which greatly reduces ...
An Empirical Study on Generalizations of the ReLU Activation Function
ACM SE '19: Proceedings of the 2019 ACM Southeast ConferenceDeep Neural Networks have become the tool of choice for Machine Learning practitioners today. They have been successfully applied for solving a large class of learning problems both in the industry and academia with applications in fields such as ...
Comments