REP: Resource-Efficient Prompting for Rehearsal-Free Continual Learning

2024-06-07

AI Summary

To address the high resource overhead and rapid accuracy degradation of prompting methods in replay-free continual learning on edge devices, this paper proposes a resource-efficient lightweight prompting framework. Our core innovations are Adaptive Token Merging (AToM) and Adaptive Layer Dropping (ALD), which jointly skip redundant computations across data and model dimensions to dynamically reduce memory footprint and computational cost. Built upon ViT/CNN backbones, the framework incorporates fast prompt selection and end-to-end differentiable prompt updating—requiring neither historical data replay nor additional parameter storage. Evaluated on three image classification benchmarks, our method achieves a 37% reduction in training memory consumption and a 42% decrease in FLOPs compared to state-of-the-art approaches, while maintaining or even improving accuracy. This effectively alleviates the longstanding trade-off between resource efficiency and model performance in edge-based continual learning.

Technology Category

• Machine Learning: Learning on the Edge & Model Compression
• Computer Vision: Diffusion Models for Vision
• Natural Language Processing: Prompt Engineering / Prompting

Application Category

• Economics, Online Markets and Human Computation: Cost models of using LLMs in production systems
• User Modeling, Personalization and Recommendation: On-Device user modeling, personalization, and recommendation
• Search and Retrieval-Augmented AI: Web learning to rank, online learning, and counterfactual learning for ranking

Abstract

Recent rehearsal-free methods, guided by prompts, generally excel in vision-related continual learning (CL) scenarios with continuously drifting data. To be deployable on real-world devices, these methods must contain high resource efficiency during training. In this paper, we introduce Resource-Efficient Prompting (REP), which targets improving the resource efficiency of prompt-based rehearsal-free methods. Our key focus is on avoiding catastrophic trade-offs with accuracy while trimming computational and memory costs during prompt learning. We achieve this by exploiting swift prompt selection that enhances input data using a carefully provisioned model, and by developing adaptive token merging (AToM) and layer dropping (ALD) algorithms for the prompt updating stage. AToM and ALD perform selective skipping across the data and model dimensions without compromising task-specific features while learning new tasks. We validate REP's superior resource efficiency over current state-of-the-art ViT- and CNN-based methods through extensive experiments on three image classification datasets.

Problem

• Continuous Learning
• Resource Efficiency
• Cueing Methods

Innovation

• Resource-Efficient Prompting
• AToM Algorithm
• ALD Algorithm