LODAP: On-Device Incremental Learning Via Lightweight Operations and Data Pruning

TL;DR

LODAP tackles on-device incremental learning for edge systems by introducing the Efficient Incremental Module (EIM), which uses intrinsic features from normal convolutions and lightweight adapters (InCFs) to learn new classes while preserving old knowledge; adapter fusion via structural re-parameterization keeps model size stable. A progressive EL2N-based data pruning strategy further reduces training data and overhead, and prototypes store compact old-class representations to replace image retention. The training objective combines new-class supervision, knowledge distillation from the old model, and prototype-based calibration, enabling strong performance with reduced computation. Experiments on CIFAR-100 and Tiny-ImageNet show up to 4.32% accuracy gains over SOTA and about 50% reductions in model complexity, with additional on-device training-time and energy savings when data pruning is used, illustrating practical applicability for EdgeAI deployments.

Abstract

Incremental learning that learns new classes over time after the model's deployment is becoming increasingly crucial, particularly for industrial edge systems, where it is difficult to communicate with a remote server to conduct computation-intensive learning. As more classes are expected to learn after their execution for edge devices. In this paper, we propose LODAP, a new on-device incremental learning framework for edge systems. The key part of LODAP is a new module, namely Efficient Incremental Module (EIM). EIM is composed of normal convolutions and lightweight operations. During incremental learning, EIM exploits some lightweight operations, called adapters, to effectively and efficiently learn features for new classes so that it can improve the accuracy of incremental learning while reducing model complexity as well as training overhead. The efficiency of LODAP is further enhanced by a data pruning strategy that significantly reduces the training data, thereby lowering the training overhead. We conducted extensive experiments on the CIFAR-100 and Tiny- ImageNet datasets. Experimental results show that LODAP improves the accuracy by up to 4.32\% over existing methods while reducing around 50\% of model complexity. In addition, evaluations on real edge systems demonstrate its applicability for on-device machine learning. The code is available at https://github.com/duanbiqing/LODAP.

Figures (6)

• Figure 1: The overview of LODAP.
• Figure 2: The accuracy variations on CIFAR-100.
• Figure 3: Classification accuracy at different incremental stages on the Tiny-ImageNet dataset.
• Figure 4: Ablation study of loss functions.
• Figure 5: Ablation study of data pruning for CIFAR-100 and Tiny-ImageNet.