PRUNIX: Non-Ideality Aware Convolutional Neural Network Pruning for Memristive Accelerators

2022/02/03 by Ali Alshaarawy, Alshaarawy, Ali, Amirali Amirsoleimani +3
Engineering · #Advanced Memory and Neural Computing #CCD and CMOS Imaging Sensors #Energy Harvesting in Wireless Networks #FOS: Computer and information sciences #Hardware Architecture (cs.AR) #Machine Learning (cs.LG)

paper · pdf · doi:10.48550/arxiv.2202.01758

openalex publication_date 2022/02/03 · openalex created_date 2022/05/05 · openalex updated_date 2026/07/28

Abstract

In this work, PRUNIX, a framework for training and pruning convolutional neural networks is proposed for deployment on memristor crossbar based accelerators. PRUNIX takes into account the numerous non-ideal effects of memristor crossbars including weight quantization, state-drift, aging and stuck-at-faults. PRUNIX utilises a novel Group Sawtooth Regularization intended to improve non-ideality tolerance as well as sparsity, and a novel Adaptive Pruning Algorithm (APA) intended to minimise accuracy loss by considering the sensitivity of different layers of a CNN to pruning. We compare our regularization and pruning methods with other standards on multiple CNN architectures, and observe an improvement of 13% test accuracy when quantization and other non-ideal effects are accounted for with an overall sparsity of 85%, which is similar to other methods

Related

• AIDX: Adaptive Inference Scheme to Mitigate State-Drift in Memristive VMM Accelerators
• Tiny but Accurate: A Pruned, Quantized and Optimized Memristor Crossbar Framework for Ultra Efficient DNN Implementation
• Adaptive Precision CNN Accelerator Using Radix-X Parallel Connected Memristor Crossbars
• TxSim:Modeling Training of Deep Neural Networks on Resistive Crossbar Systems
• GENIEx: A Generalized Approach to Emulating Non-Ideality in Memristive\n Xbars using Neural Networks