Class-Incremental Learning with Parameter-Efficient Cross-Task Prompts

qian feng, Hanbin Zhao, Chao Zhang, Hui QianOpenReview 2024 · Submission to ICLR 2024 · Rejected · OpenReview

Abstract

Class-Incremental Learning (CIL) aims to learn deep models on sequential tasks continually, where each new task includes a batch of new classes and deep models do not have access to task-ID information at the inference time. Recent vast pre-trained models (PTMs) have achieved outstanding performance by prompt technique in practical CIL without the old samples (rehearsal-free) and with a memory constraint (memory-constrained): Prompt-extending and Prompt-fixed methods. However, prompt-extending methods need a large memory buffer to maintain an ever-expanding prompt pool and meet an extra challenging prompt selection problem. Prompt-fixed methods only learn a fixed number of prompts on one of the incremental tasks and can not handle all the incremental tasks effectively. To achieve a good balance between the memory cost and the performance on all the tasks, we propose a Parameter-Efficient Cross-Task Prompt (PECTP) framework with a prompt retention module (PRM). To make the final learned prompts effective on the whole incremental tasks, PRM constrains the evolution of cross-task prompts' parameters from Outer Prompt Granularity and Inner Prompt Granularity. Extensive experiments show the effectiveness of our method.
BibTeX
@misc{feng2024class,
  author = {qian feng and Hanbin Zhao and Chao Zhang and Hui Qian},
  title = {{Class-Incremental Learning with Parameter-Efficient Cross-Task Prompts}},
  year = {2024},
  url = {https://openreview.net/forum?id=4lqo5Jwfnq},
  howpublished = {OpenReview},
  note = {Submitted to ICLR 2024},
}
Figure