HAWX: A Hardware-Aware FrameWork for Fast and Scalable ApproXimation of DNNs

2026-02-18

AI Summary

This work addresses the challenge of efficiently exploring the vast design space of approximate deep neural networks (DNNs), where exhaustive search becomes intractable due to poor scalability and suboptimal trade-offs among accuracy, power, and area. To overcome this, the authors propose HAWX, a framework that introduces the first multi-granularity sensitivity analysis spanning operators, filters, layers, and entire models. HAWX integrates hardware-aware search algorithms with predictive models for accuracy, power, and area to guide the selective integration of heterogeneous approximate computing (AxC) modules. It supports spatial-temporal accelerator architectures and accommodates both off-the-shelf and custom approximate components, achieving exponentially increasing search efficiency with model scale. Experiments demonstrate 23× and 3 million× speedups for layer- and filter-level searches on LeNet-5, respectively, matching exhaustive search accuracy, while also validating scalability on VGG-11, ResNet-18, and EfficientNetLite.

Technology Category

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

Application Category

• Search and Retrieval-Augmented AI: Efficiency and scalability of Web search engines
• Graph Algorithms and Modeling for the Web: Graph neural networks and deep learning approaches for Web-related graphs
• Systems and Infrastructure for Web, Mobile and WoT: Applied ML and AI for Web-based mobile applications

Abstract

This work presents HAWX, a hardware-aware scalable exploration framework that employs multi-level sensitivity scoring at different DNN abstraction levels (operator, filter, layer, and model) to guide selective integration of heterogeneous AxC blocks. Supported by predictive models for accuracy, power, and area, HAWX accelerates the evaluation of candidate configurations, achieving over 23* speedup in a layer-level search with two candidate approximate blocks and more than (3*106)* speedup at the filter-level search only for LeNet-5, while maintaining accuracy comparable to exhaustive search. Experiments across state-of-the-art DNN benchmarks such as VGG-11, ResNet-18, and EfficientNetLite demonstrate that the efficiency benefits of HAWX scale exponentially with network size. The HAWX hardware-aware search algorithm supports both spatial and temporal accelerator architectures, leveraging either off-the-shelf approximate components or customized designs.

Problem

• approximate computing
• hardware-aware optimization
• DNN approximation
• scalable exploration
• accuracy-efficiency tradeoff

Innovation

• hardware-aware approximation
• multi-level sensitivity scoring
• approximate computing (AxC)
• scalable DNN optimization
• predictive modeling