Lightweight Neural Network Architectures for Efficient Artificial Intelligence Processing on Edge Devices

Aoife Brennan

Lightweight Neural Network Architectures for Efficient Artificial Intelligence Processing on Edge Devices

Author(s)	Aoife Brennan
Country	Ireland
Abstract	Artificial intelligence applications are increasingly expected to operate directly on smartphones, embedded controllers, sensor platforms, robots, and other edge devices. Conventional deep neural networks can provide strong predictive performance but often require substantial memory, computation, and energy. These requirements can make cloud-independent inference difficult on hardware with restricted processing capacity. Lightweight neural-network design therefore focuses on reducing parameters and arithmetic operations while retaining sufficient predictive accuracy for practical on-device artificial intelligence. This hypothetical methodological study compares alternative lightweight convolutional architectures for image classification under an edge-computing scenario. A fictional dataset containing 12,000 images across twelve recognition categories is divided into training, validation, and testing subsets. Four conceptual architectures are examined: a conventional convolutional network, a SqueezeNet-style compact model, a MobileNet-style depthwise-separable network, and a ShuffleNet-style grouped-convolution architecture. Models are evaluated using classification accuracy, parameter count, model size, multiply-accumulate operations, and hypothetical inference latency. The illustrative analysis indicates that the conventional network achieves competitive classification accuracy but requires substantially greater storage and computation. The MobileNet-style model provides the strongest hypothetical accuracy–latency balance, whereas the ShuffleNet-style architecture records the lowest latency and computational requirement. A second analysis examines 8-bit quantization and structured pruning of the MobileNet-style model and demonstrates further reductions in model size, latency, and estimated inference energy with a limited illustrative reduction in classification accuracy. The study concludes that efficient edge artificial intelligence requires joint consideration of predictive accuracy, computational operations, memory consumption, latency, and energy rather than optimization of accuracy alone. Architecture-level efficiency and post-training compression can provide complementary benefits. However, actual deployment performance depends on hardware architecture, runtime implementation, memory access, operator support, thermal constraints, and application-specific latency requirements.
Keywords	Edge Artificial Intelligence, Lightweight Neural Networks, MobileNet, SqueezeNet, ShuffleNet, Model Compression, Quantization, Embedded AI, Efficient Deep Learning, Edge Computing.
Field	Engineering
Published In	Volume 2, Issue 4, July-August 2026
Published On	2026-08-12

View / Download PDF File

E-ISSN: 3067-5073

doi

CrossRef DOI is assigned to each research paper published in our journal.

IJMLAIDSE DOI prefix is
10.00000/IJMLAIDSE

Downloads

Research Paper Format Copyright Permission and Undertaking Form

All research papers published on this website are licensed under Creative Commons Attribution-ShareAlike 4.0 International License, and all rights belong to their respective authors/researchers.

CC-BY-SA

About IJMLAIDSE Fees & Payment Current Issue Publication Archive	Submit Research Paper Track Submission Status Publication Guidelines Publication Ethics Peer Review & Plagiarism	Join as a Reviewer Editors & Reviewers Get Reviewer Membership Certi.	Website/Journal Policies Usage Policy Content Policies Privacy Policy

Contact Us	+1-872-302-3851	editor@uscse.com

International Journal of Machine Learning, AI & Data Science Evolution

A Widely Indexed Open Access Peer Reviewed Multidisciplinary Monthly Scholarly International Journal

Lightweight Neural Network Architectures for Efficient Artificial Intelligence Processing on Edge Devices

Share this