Ex Situ Transfer of Bayesian Neural Networks to Resistive Memory‐Based Inference Hardware. (20th May 2021)

• Record Type: Journal Article
• Authors: Dalgaty, Thomas; Esmanhotto, Eduardo; Castellani, Niccolo; Querlioz, Damien; Vianello, Elisa
• Is Part Of: Advanced intelligent systems. Volume 3:Number 8(2021)
• Page Start: n/a
• Page End: n/a
• Publication Date: 2021-05-20
• Subjects: Bayesian machine learning -- memristors -- neural networks -- resistive memories; Artificial intelligence -- Periodicals; Robotics -- Periodicals; Control theory -- Periodicals; 006.3
• DOI: 10.1002/aisy.202000103
• Languages: English
• ISSNs: 2640-4567

Abstract

Abstract : Neural networks cannot typically be trained locally in edge‐computing systems due to severe energy constraints. It has, therefore, become commonplace to train them "ex situ" and transfer the resulting model to a dedicated inference hardware. Resistive memory arrays are of particular interest for realizing such inference hardware, because they offer an extremely low‐power implementation of the dot‐product operation. However, the transfer of high‐precision software parameters to the imprecise and random conductance states of resistive memories poses significant challenges. Here, it is proposed that Bayesian neural networks can be more suitable for model transfer, because, such as device conductance states, their parameters are described by random variables. The ex situ training of a Bayesian neural network is performed, and then, the resulting software model is transferred in a single programming step to an array of 16 384 resistive memory devices. On an illustrative classification task, it is observed that the transferred decision boundaries and the prediction uncertainties of the software model are well preserved. This work demonstrates that resistive memory‐based Bayesian neural networks are a promising direction in the development of resistive memory compatible edge inference hardware. Abstract : It is experimentally demonstrated how resistive memory‐based edge inference can be achieved using Bayesian neural networks. Since, like resistive memory devices, Bayesian network parameters are random variables, a more natural pairing of device and algorithm is proposed. It is also discussed how prediction uncertainty, available via the Bayesian approach, can be indispensable in safety‐critical edge applications.