Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Equivariant-Aware Structured Pruning for Efficient Edge Deployment: A Comprehensive Framework with Adaptive Fine-Tuning

Mohammed Alnemari

TL;DR

This work tackles the challenge of deploying efficient, transformation-invariant models on edge devices by uniting $C_4$-equivariant CNNs (via the e2cnn library) with an equivariance-aware structured pruning framework. It introduces a layer-type-aware pruning strategy that preserves equivariant layers while compressing linear components, coupled with adaptive fine-tuning and INT8 quantization to recover accuracy and reach deployment readiness. The method achieves substantial compression (e.g., 29.3% through distillation and up to 87.6% total parameter reduction after full optimization) with minimal or even improved accuracy on rotated data, demonstrating preserved geometric robustness. The resulting pipeline provides a reproducible, performance-oriented approach that bridges group-theoretic network design and practical edge deployment, with particular promise for satellite imagery and other geometry-rich vision tasks.

Abstract

This paper presents a novel framework combining group equivariant convolutional neural networks (G-CNNs) with equivariant-aware structured pruning to produce compact, transformation-invariant models for resource-constrained environments. Equivariance to rotations is achieved through the C4 cyclic group via the e2cnn library,enabling consistent performance under geometric transformations while reducing computational overhead. Our approach introduces structured pruning that preserves equivariant properties by analyzing e2cnn layer structure and applying neuron-level pruning to fully connected components. To mitigate accuracy degradation, we implement adaptive fine-tuning that automatically triggers when accuracy drop exceeds 2%, using early stopping and learning rate scheduling for efficient recovery. The framework includes dynamic INT8 quantization and a comprehensive pipeline encompassing training, knowledge distillation, structured pruning, fine-tuning, and quantization. We evaluate our method on satellite imagery (EuroSAT) and standard benchmarks (CIFAR-10, Rotated MNIST) demonstrating effectiveness across diverse domains. Experimental results show 29.3% parameter reduction with significant accuracy recovery, demonstrating that structured pruning of equivariant networks achieves substantial compression while maintaining geometric robustness. Our pipeline provides a reproducible framework for optimizing equivariant models, bridging the gap between group-theoretic network design and practical deployment constraints, with particular relevance to satellite imagery analysis and geometric vision tasks.

Equivariant-Aware Structured Pruning for Efficient Edge Deployment: A Comprehensive Framework with Adaptive Fine-Tuning

TL;DR

This work tackles the challenge of deploying efficient, transformation-invariant models on edge devices by uniting -equivariant CNNs (via the e2cnn library) with an equivariance-aware structured pruning framework. It introduces a layer-type-aware pruning strategy that preserves equivariant layers while compressing linear components, coupled with adaptive fine-tuning and INT8 quantization to recover accuracy and reach deployment readiness. The method achieves substantial compression (e.g., 29.3% through distillation and up to 87.6% total parameter reduction after full optimization) with minimal or even improved accuracy on rotated data, demonstrating preserved geometric robustness. The resulting pipeline provides a reproducible, performance-oriented approach that bridges group-theoretic network design and practical edge deployment, with particular promise for satellite imagery and other geometry-rich vision tasks.

Abstract

This paper presents a novel framework combining group equivariant convolutional neural networks (G-CNNs) with equivariant-aware structured pruning to produce compact, transformation-invariant models for resource-constrained environments. Equivariance to rotations is achieved through the C4 cyclic group via the e2cnn library,enabling consistent performance under geometric transformations while reducing computational overhead. Our approach introduces structured pruning that preserves equivariant properties by analyzing e2cnn layer structure and applying neuron-level pruning to fully connected components. To mitigate accuracy degradation, we implement adaptive fine-tuning that automatically triggers when accuracy drop exceeds 2%, using early stopping and learning rate scheduling for efficient recovery. The framework includes dynamic INT8 quantization and a comprehensive pipeline encompassing training, knowledge distillation, structured pruning, fine-tuning, and quantization. We evaluate our method on satellite imagery (EuroSAT) and standard benchmarks (CIFAR-10, Rotated MNIST) demonstrating effectiveness across diverse domains. Experimental results show 29.3% parameter reduction with significant accuracy recovery, demonstrating that structured pruning of equivariant networks achieves substantial compression while maintaining geometric robustness. Our pipeline provides a reproducible framework for optimizing equivariant models, bridging the gap between group-theoretic network design and practical deployment constraints, with particular relevance to satellite imagery analysis and geometric vision tasks.

Paper Structure

This paper contains 26 sections, 4 equations, 1 figure, 5 tables, 1 algorithm.

Table of Contents

  1. Introduction
  2. Background
  3. Edge Computing and Model Compression
  4. Geometric Robustness in Efficient Models
  5. Equivariant Model Compression
  6. Related Work
  7. Equivariant Convolutional Networks
  8. Structured Pruning
  9. Compression for Edge AI
  10. Methodology
  11. Group Equivariant Convolution
  12. Equivariance-Aware Structured Pruning
  13. Layer-Type-Aware Pruning Algorithm
  14. Quantization and Deployment Preparation
  15. Pipeline Overview
  16. ...and 11 more sections

Figures (1)

  • Figure 1: Our equivariance-aware optimization pipeline: training, distillation, selective pruning, adaptive fine-tuning, and quantization for deployment preparation.