PlatformX: An End-to-End Transferable Platform for Energy-Efficient Neural Architecture Search

2025-10-10

AI Summary

Existing hardware-aware neural architecture search (HW-NAS) methods suffer from poor cross-platform portability, low energy estimation accuracy, high search overhead, and reliance on manual analysis. This paper proposes an end-to-end, transferable hardware-aware NAS framework for automated, energy-efficient DNN design on edge devices. Key contributions include: (1) an energy-driven compact search space; (2) a transferable energy prediction model based on kernel-level hardware features; (3) a Pareto-driven multi-objective efficient search algorithm; and (4) a high-accuracy runtime energy profiling system integrated with external power monitoring. Evaluated across multiple mobile platforms, the framework significantly reduces search cost—requiring several-fold fewer architecture evaluations—while yielding models achieving 0.94 Top-1 accuracy or as low as 0.16 mJ/inference on ImageNet, consistently outperforming MobileNet-V2.

Technology Category

• Machine Learning: Hardware-aware ML
• Search and Optimization: Learning to Search
• Computer Vision: Learning & Optimization for CV

Application Category

• Systems and Infrastructure for Web, Mobile and WoT: Energy management for devices in mobile Web and WoT environments
• Search and Retrieval-Augmented AI: Efficiency and scalability of Web search engines
• Economics, Online Markets and Human Computation: Incentives in network design for Web infrastructures and ecosystems

Abstract

Hardware-Aware Neural Architecture Search (HW-NAS) has emerged as a powerful tool for designing efficient deep neural networks (DNNs) tailored to edge devices. However, existing methods remain largely impractical for real-world deployment due to their high time cost, extensive manual profiling, and poor scalability across diverse hardware platforms with complex, device-specific energy behavior. In this paper, we present PlatformX, a fully automated and transferable HW-NAS framework designed to overcome these limitations. PlatformX integrates four key components: (i) an energy-driven search space that expands conventional NAS design by incorporating energy-critical configurations, enabling exploration of high-efficiency architectures; (ii) a transferable kernel-level energy predictor across devices and incrementally refined with minimal on-device samples; (iii) a Pareto-based multi-objective search algorithm that balances energy and accuracy to identify optimal trade-offs; and (iv) a high-resolution runtime energy profiling system that automates on-device power measurement using external monitors without human intervention. We evaluate PlatformX across multiple mobile platforms, showing that it significantly reduces search overhead while preserving accuracy and energy fidelity. It identifies models with up to 0.94 accuracy or as little as 0.16 mJ per inference, both outperforming MobileNet-V2 in accuracy and efficiency. Code and tutorials are available at github.com/amai-gsu/PlatformX.

Problem

• Automating energy-efficient neural architecture search for edge devices
• Overcoming high time cost and poor hardware scalability limitations
• Enabling transferable energy prediction across diverse hardware platforms

Innovation

• Energy-driven search space for efficient architectures
• Transferable kernel-level energy predictor across devices
• Pareto-based multi-objective search balancing energy and accuracy