Efficient Deep Learning for Biometrics: Overview, Challenges and Trends in Ear of Frugal AI

TL;DR

The paper tackles the high computational and energy demands of deep learning in biometric applications by surveying Efficient Deep Learning (EDL) techniques designed for edge and resource-constrained settings. It provides a taxonomy across data-centric, model-centric, and deployment-centric methods, and outlines a multi-metric evaluation framework including $FLOPs/MACs$, memory, latency, throughput, and energy to compare approaches fairly. The review covers state-of-the-art data augmentation, distillation, quantization, pruning, and hardware-aware deployment, and discusses biometric-specific applications such as on-device authentication, edge surveillance, liveness detection, and privacy-preserving templates. It highlights challenges in accuracy-efficiency trade-offs, generalization versus specialization, security, benchmarking standardization, and hardware-software co-design, and suggests directions like dynamic adaptation, neuro-symbolic AI, automated efficient design, and fairness considerations to guide future work. Together, these insights aim to enable practical, scalable, and secure biometric systems that operate effectively on edge devices while maintaining privacy and robustness.

Abstract

Recent advances in deep learning, whether on discriminative or generative tasks have been beneficial for various applications, among which security and defense. However, their increasing computational demands during training and deployment translates directly into high energy consumption. As a consequence, this induces a heavy carbon footprint which hinders their widespread use and scalability, but also a limitation when deployed on resource-constrained edge devices for real-time use. In this paper, we briefly survey efficient deep learning methods for biometric applications. Specifically, we tackle the challenges one might incur when training and deploying deep learning approaches, and provide a taxonomy of the various efficient deep learning families. Additionally, we discuss complementary metrics for evaluating the efficiency of these models such as memory, computation, latency, throughput, and advocate for universal and reproducible metrics for better comparison. Last, we give future research directions to consider.

Figures (2)

• Figure 1: Overview taxonomy of Efficient Deep Learning (EDL)
• Figure 2: Overview diagram of three model-centric techniques. (a) Knowledge distillation, (b) quantization, (c) pruning.