MMA: A Momentum Mamba Architecture for Human Activity Recognition with Inertial Sensors

2025-11-26

AI Summary

Existing deep models (CNNs, RNNs, Transformers) for inertial-sensor-based human activity recognition (HAR) suffer from gradient instability, high computational overhead, and insufficient modeling of long-range temporal dependencies. To address these issues, we propose Complex Momentum Mamba (CMM), a momentum-enhanced structured state space model. CMM is the first to incorporate a second-order momentum mechanism into the Mamba architecture, enabling stable long-term memory retention. Furthermore, it introduces a complex-domain frequency-selective memory scaling mechanism to achieve frequency-aware memory modulation. Retaining linear-time complexity, CMM significantly accelerates convergence and enhances robustness in long-sequence modeling. On multiple HAR benchmarks, CMM consistently outperforms standard Mamba and Transformer baselines in accuracy, robustness, and training efficiency—achieving optimal performance-complexity trade-offs under moderate computational cost.

Technology Category

• Machine Learning: Large Multimodal Models (LMMs)
• Cognitive Modeling & Cognitive Systems: Neural Spike Coding
• Planning, Routing, and Scheduling: Activity and Plan Recognition

Application Category

• Graph Algorithms and Modeling for the Web: Foundation models and LLMs for Web-related graphs
• User Modeling, Personalization and Recommendation: On-Device user modeling, personalization, and recommendation
• Semantics and Knowledge: Methods to enhance, augment, integrate or synergize semantic models such as knowledge graphs and LLMs

Abstract

Human activity recognition (HAR) from inertial sensors is essential for ubiquitous computing, mobile health, and ambient intelligence. Conventional deep models such as Convolutional Neural Networks (CNNs), Recurrent Neural Networks (RNNs), and transformers have advanced HAR but remain limited by vanishing or exloding gradients, high computational cost, and difficulty in capturing long-range dependencies. Structured state-space models (SSMs) like Mamba address these challenges with linear complexity and effective temporal modeling, yet they are restricted to first-order dynamics without stable longterm memory mechanisms. We introduce Momentum Mamba, a momentum-augmented SSM that incorporates second-order dynamics to improve stability of information flow across time steps, robustness, and long-sequence modeling. Two extensions further expand its capacity: Complex Momentum Mamba for frequency-selective memory scaling. Experiments on multiple HAR benchmarks demonstrate consistent gains over vanilla Mamba and Transformer baselines in accuracy, robustness, and convergence speed. With only moderate increases in training cost, momentum-augmented SSMs offer a favorable accuracy-efficiency balance, establishing them as a scalable paradigm for HAR and a promising principal framework for broader sequence modeling applications.

Problem

• Improving human activity recognition using inertial sensors with enhanced temporal modeling
• Addressing vanishing gradients and long-range dependency issues in sequence models
• Developing efficient architectures with better accuracy-efficiency balance for HAR

Innovation

• Momentum-augmented SSM with second-order dynamics
• Complex Momentum Mamba for frequency-selective scaling
• Linear complexity with improved long-sequence modeling