Astromorphic Self-Repair of Neuromorphic Hardware Systems

Authors: Zhuangyu Han, A N M Nafiul Islam, Abhronil Sengupta

AAAI 2023 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental Performance evaluation of the proposed learning rule is demonstrated for an unsupervised learning framework on the MNIST and F-MNIST datasets. ... Table 1 reports the best accuracy recorded in the self-repair process for hardware-realistic faults in the network (stuck-at-faults and weight drift). The results have been averaged over 5 independent runs of the network.
Researcher Affiliation Academia Zhuangyu Han, A N M Nafiul Islam, Abhronil Sengupta School of Electrical Engineering and Computer Science Penn State University University Park, PA 16802 zfh5141@psu.edu, nafiul@psu.edu, sengupta@psu.edu
Pseudocode No The paper describes algorithms and formulations using mathematical equations and text, but it does not present any structured pseudocode or algorithm blocks.
Open Source Code Yes Our implementation source code and trained models are available at https://github.com/ Neuro Comp Lab-psu/Astromorphic Self Repair.
Open Datasets Yes The SNNs are trained using STDP on two recognition tasks, namely the MNIST (Le Cun, Cortes, and Burges 2010) and Fashion MNIST (Xiao, Rasul, and Vollgraf 2017) datasets.
Dataset Splits No The paper mentions training on datasets but does not explicitly detail validation splits. It focuses on training and then injecting faults for self-repair evaluation. While common datasets like MNIST often have standard splits, the paper doesn't specify them for its experimental setup beyond training.
Hardware Specification Yes The algorithms were run on the hardware environment consisting of one Intel(R) Xeon(R) Silver 4210 CPU, one NVIDIA Ge Force RTX 2080 Ti GPU with 11264 MBytes graphics memory, 187 GBytes RAM and Cent OS Linux 7 operating system.
Software Dependencies Yes The proposed A-STDP (local) learning rule is implemented using the Binds NET (Hazan et al. 2018) framework an open-source SNN simulation platform based on Py Torch (https://pytorch.org/). The text mentions PyTorch which implies a version, even if not explicitly stated with a number.
Experiment Setup Yes The detailed information regarding all simulation hyperparameters is included in Table 2 and typical ablation studies for important hyperparameters of the astromorphic learning model are also included in the supplementary material (Han, Islam, and Sengupta 2022).