On Accelerating Edge AI: Optimizing Resource-Constrained Environments

TL;DR

The paper surveys strategies for accelerating AI on resource-constrained edge devices, focusing on three pillars: model compression (quantization, pruning, tensor decomposition, distillation), neural architecture search (NAS), and compiler/deployment frameworks. It reviews methods to tailor architectures and representations to hardware budgets, discusses practical deployment pipelines, and presents benchmarks and case studies that integrate compression and NAS with hardware-aware constraints. The work highlights open challenges such as pre-training pruning for massive models, loss-topology implications in distillation, and scalability of NAS for very large models, while proposing a path toward a unified, model-agnostic optimization framework for edge AI. The findings emphasize that combining compression, NAS, and compiler optimizations can yield scalable, platform-independent edge AI solutions with significant reductions in latency, memory, and energy usage, enabling practical real-time deployment.

Abstract

Resource-constrained edge deployments demand AI solutions that balance high performance with stringent compute, memory, and energy limitations. In this survey, we present a comprehensive overview of the primary strategies for accelerating deep learning models under such constraints. First, we examine model compression techniques-pruning, quantization, tensor decomposition, and knowledge distillation-that streamline large models into smaller, faster, and more efficient variants. Next, we explore Neural Architecture Search (NAS), a class of automated methods that discover architectures inherently optimized for particular tasks and hardware budgets. We then discuss compiler and deployment frameworks, such as TVM, TensorRT, and OpenVINO, which provide hardware-tailored optimizations at inference time. By integrating these three pillars into unified pipelines, practitioners can achieve multi-objective goals, including latency reduction, memory savings, and energy efficiency-all while maintaining competitive accuracy. We also highlight emerging frontiers in hierarchical NAS, neurosymbolic approaches, and advanced distillation tailored to large language models, underscoring open challenges like pre-training pruning for massive networks. Our survey offers practical insights, identifies current research gaps, and outlines promising directions for building scalable, platform-independent frameworks to accelerate deep learning models at the edge.

Figures (12)

• Figure 1: Strategies and Tools for Enhancing AI Edge Computing
• Figure 2: Pruning Taxonomy as formulated by pruning_survey_cheng
• Figure 3: Kernel approximation of low-rank decomposition matrix as depicted in model_compression_survey
• Figure 4: Distillation Components
• Figure 5: Concepts and Tools in Neural Architecture Search