Summary of Safe: Slow and Fast Parameter-efficient Tuning For Continual Learning with Pre-trained Models, by Linglan Zhao et al.
SAFE: Slow and Fast Parameter-Efficient Tuning for Continual Learning with Pre-Trained Models
by Linglan Zhao, Xuerui Zhang, Ke Yan, Shouhong Ding, Weiran Huang
First submitted to arxiv on: 4 Nov 2024
Categories
- Main: Machine Learning (cs.LG)
- Secondary: Computer Vision and Pattern Recognition (cs.CV)
GrooveSquid.com Paper Summaries
GrooveSquid.com’s goal is to make artificial intelligence research accessible by summarizing AI papers in simpler terms. Each summary below covers the same AI paper, written at different levels of difficulty. The medium difficulty and low difficulty versions are original summaries written by GrooveSquid.com, while the high difficulty version is the paper’s original abstract. Feel free to learn from the version that suits you best!
| Summary difficulty | Written by | Summary |
|---|---|---|
| High | Paper authors | High Difficulty Summary Read the original abstract here |
| Medium | GrooveSquid.com (original content) | Medium Difficulty Summary The paper proposes a framework for continual learning, which aims to incrementally acquire new concepts while retaining previous knowledge. With powerful pre-trained models (PTMs) becoming increasingly popular, researchers are exploring ways to train incremental learning systems using these foundation models rather than starting from scratch. Existing works often view PTMs as a strong initial point and apply parameter-efficient tuning (PET) in the first session for adapting to downstream tasks, but then freeze model parameters to tackle forgetting issues. The proposed Slow And Fast parameter-Efficient tuning (SAFE) framework addresses these limitations by incorporating a transfer loss function to inherit general knowledge from foundation models. This is achieved by calibrating slow efficient tuning parameters in the first session and continuously updating fast ones. An entropy-based aggregation strategy is also introduced for dynamic utilization of complementarity between slow and fast learners during inference. The paper demonstrates the effectiveness of SAFE on seven benchmark datasets, significantly surpassing state-of-the-art methods. |
| Low | GrooveSquid.com (original content) | Low Difficulty Summary This paper is about making machines learn new things without forgetting what they already know. Researchers are trying to use powerful starting points (pre-trained models) to help machines learn more effectively. The problem with current approaches is that they either don’t fully utilize the power of these starting points or forget important information along the way. The proposed solution, called SAFE, addresses this issue by finding a balance between retaining old knowledge and acquiring new skills. It’s like having two different “brains” in one machine: one for remembering what it already knows and another for learning new things. This approach was tested on several datasets and performed better than previous methods. |
Keywords
» Artificial intelligence » Continual learning » Inference » Loss function » Parameter efficient
