Summary of Integrating Dual Prototypes For Task-wise Adaption in Pre-trained Model-based Class-incremental Learning, by Zhiming Xu et al.
Integrating Dual Prototypes for Task-Wise Adaption in Pre-Trained Model-Based Class-Incremental Learning
by Zhiming Xu, Suorong Yang, Baile Xu, Jian Zhao, Furao Shen
First submitted to arxiv on: 26 Nov 2024
Categories
- Main: Machine Learning (cs.LG)
- Secondary: Machine Learning (stat.ML)
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 Dual Prototype network for Task-wise Adaption (DPTA) is a novel approach to class-incremental learning, which aims to acquire new classes while preserving historical knowledge incrementally. This method leverages pre-trained models and fine-tunes them on downstream incremental tasks using task streams. To prevent catastrophic forgetting, DPTA incorporates a center-adapt loss that forces the representation to be more centrally clustered and class separable. The network also enables test-time adapter selection, utilizing raw prototypes to deduce possible task indexes and augmented prototypes to determine the final result. Experimental results on several benchmark datasets demonstrate the state-of-the-art performance of DPTA. |
| Low | GrooveSquid.com (original content) | Low Difficulty Summary Class-incremental learning is a way for machines to learn new classes while keeping old knowledge. This helps in real-life situations where data comes in chunks, like new animal species being discovered. Some methods use pre-trained models that are fine-tuned on the new information. However, this can cause “forgetting” of old knowledge if not done carefully. The DPTA (Dual Prototype network for Task-wise Adaption) is a new way to do this without forgetting too much. It uses special losses and selections to make sure the model learns from both the old and new data. |
