COMponent-Aware Pruning for Accelerated Control Tasks in Latent Space Models

2025-08-11

AI Summary

Deploying neural network controllers (NNCs) on resource-constrained mobile platforms—such as mobile robots, wearables, and IoT endpoints—is hindered by the high computational and memory demands of deep neural networks (DNNs), which compromise real-time performance and closed-loop stability. To address this, we propose a component-aware structured pruning method that explicitly incorporates control-theoretic semantics into the pruning process and integrates Lyapunov stability analysis. This enables theoretically grounded, safety-certified model compression by establishing rigorous bounds linking sparsity level to closed-loop stability. Our approach unifies temporal-difference model predictive control (TD-MPC) with latent-space control component optimization, significantly reducing parameter count and inference latency while strictly preserving control performance and system stability. Experimental results quantify the maximum safe compression ratio, providing both theoretical guarantees and practical deployment strategies for reliable edge-based NNCs.

Technology Category

• Machine Learning: Learning on the Edge & Model Compression
• Natural Language Processing: Safety and Robustness
• Intelligent Robots: Behavior Learning & Control

Application Category

• Responsible Web: Machine-in-the-loop, human agency and autonomy
• Security and Privacy: Security and privacy of machine learning and AI applications
• Systems and Infrastructure for Web, Mobile and WoT: Applied ML and AI for Web-based mobile applications

Abstract

The rapid growth of resource-constrained mobile platforms, including mobile robots, wearable systems, and Internet-of-Things devices, has increased the demand for computationally efficient neural network controllers (NNCs) that can operate within strict hardware limitations. While deep neural networks (DNNs) demonstrate superior performance in control applications, their substantial computational complexity and memory requirements present significant barriers to practical deployment on edge devices. This paper introduces a comprehensive model compression methodology that leverages component-aware structured pruning to determine the optimal pruning magnitude for each pruning group, ensuring a balance between compression and stability for NNC deployment. Our approach is rigorously evaluated on Temporal Difference Model Predictive Control (TD-MPC), a state-of-the-art model-based reinforcement learning algorithm, with a systematic integration of mathematical stability guarantee properties, specifically Lyapunov criteria. The key contribution of this work lies in providing a principled framework for determining the theoretical limits of model compression while preserving controller stability. Experimental validation demonstrates that our methodology successfully reduces model complexity while maintaining requisite control performance and stability characteristics. Furthermore, our approach establishes a quantitative boundary for safe compression ratios, enabling practitioners to systematically determine the maximum permissible model reduction before violating critical stability properties, thereby facilitating the confident deployment of compressed NNCs in resource-limited environments.

Problem

• Optimize neural network pruning for resource-constrained mobile platforms
• Balance compression and stability in neural network controllers
• Determine safe compression limits while preserving control stability

Innovation

• Component-aware pruning optimizes compression and stability
• Integrates Lyapunov criteria for stability guarantees
• Quantitative boundary for safe compression ratios