Skip to content

Optimizing MNIST Digit Recognition for Edge Devices: A Multi-dimensional Evaluation Approach

2026 · Journal of Ad-hoc Network and Mobile Computing · 0 citations

TL;DR

The authors introduce the Edge Suitability Score (ESS), a composite metric that combines normalized accuracy, model size, and inference speed into a single value, weighted at 0.40, 0.35, and 0.25 to reflect their relative importance for microcontroller deployment.

Abstract

Handwritten digit recognition on the MNIST dataset routinely achieves near-perfect accuracy. However, deploying these models on edge platforms introduces new challenges where predictive accuracy is no longer the only metric that matters. For constrained systems, memory footprint, computational overhead, and inference latency are equally critical. In this work, they propose a multi-dimensional evaluation framework for deployment-oriented assessment. The authors introduce the Edge Suitability Score (ESS), a composite metric that combines normalized accuracy, model size, and inference speed into a single value, weighted at 0.40, 0.35, and 0.25, respectively, to reflect their relative importance for microcontroller deployment. By comparing two lightweight architectures, a scaled-down CNN (L-CNN) and a depthwise-separable L-MobileNet, against a deeper Baseline CNN, the results show that compact networks can maintain near-99% accuracy while drastically reducing storage and computation requirements: L-MobileNet achieves 99.10% accuracy with only 12,186 parameters and roughly 48 KB of weight memory, compared with 99.45% accuracy and over 1 MB for the baseline. This framework offers a practical methodology for selecting neural networks in real-world edge environments, bridging the gap between theoretical performance and actual deployability on resource-constrained hardware such as the STM32 and ESP32.

View source

Similar papers

Open access Jul 2026

Optimizing 1001-Class Handwritten Digit Sequence Recognition for the Mexican Electoral Process Using Asymmetric ResNet-18 and CBAM

A specialized Handwritten Digit Sequence Recognition (HDSR) framework for the Mexican Preliminary Election Results Program (PREP) based on a modified ResNet-18 architecture is proposed, introducing an asymmetric stride designed explicitly to preserve the 1:3 horizontal feature resolution of electoral tally sheets.

Miguel Ángel Camargo-Rojas, G. Sánchez-Pérez, J. Portillo-Portillo et al. · 0 citations
Aug 2026

A Dynamic Stochastic Computing-Based Approach to Edge Inference for Convolutional Neural Networks

A comparative analysis of dynamic stochastic computing using a convolutional neural network for MNIST digit classification indicates that dynamic stochastic computing can serve as an efficient alternative for the design and implementation of neural-network accelerators.

G. Rajashekhar, P. Muralidhar · 0 citations
Preprint Jul 2026

Barnamala: Parameter-Efficient Handwritten Devanagari Recognition at Benchmark Saturation

We built a compact convolutional network (1.11 M parameters) for 46-class DHCD Devanagari recognition and reached 99.73%, the highest reported at 15.6x smaller than prior state-of-the-art. We have effectively reached the saturation point: every model tested, large teacher ensembles included, hits the same 11-error intrinsic floor. No configuration achieves a statistically clear win under exact McNemar tests with Wilson confidence intervals. Even without knowledge distillation, our student matches the nearest large-model baseline (17.32 M parameters; McNemar $p = 0.345$). Outside of DHCD, zero-shot on CMATERdb digits gives 76.6% and fine-tuning reaches 97.8%; corruption robustness is also far better than large baselines (mean corruption accuracy 75.7% vs. 38.7%). All artifacts are at https://github.com/Ampixa/barnamala.

Ashish Thapa, Samrat Karki · 0 citations
Open access 2020

A Comparative Study of Deep Learning Architectures for Image Classification

Image classification is a major area in computer vision, driven by rapid advances in deep learning. Over the last decade, convolutional neural networks (CNNs) and their variants have achieved high performance in applications such as medical diagnosis, autonomous driving, industrial inspection, remote sensing, and biometrics. However, choosing the right model remains challenging due to trade-offs between accuracy, computational cost, efficiency, and robustness. This paper presents a comparative study of different deep learning architectures, including classical CNNs, deep hierarchical models, residual and dense networks, and compound-scaled architectures. Using a common evaluation framework and standard datasets, the study analyzes performance based on key design factors such as depth, width, receptive field, skip connections, and normalization. Theoretical concepts like convolution operations, residual learning, and optimization are also discussed.The results show that deeper networks provide better representation, while residual connections and compound scaling improve training stability and efficiency. Lightweight models perform well in resource-limited and real-time environments. Overall, the study offers practical guidance for selecting suitable architectures and highlights future research areas such as neural architecture search, self-supervised learning, and efficient model deployment.

Riyaz Mohammed · 0 citations