Summary of Lpt++: Efficient Training on Mixture Of Long-tailed Experts, by Bowen Dong et al.
LPT++: Efficient Training on Mixture of Long-tailed Experts
by Bowen Dong, Pan Zhou, Wangmeng Zuo
First submitted to arxiv on: 17 Sep 2024
Categories
- Main: Computer Vision and Pattern Recognition (cs.CV)
- Secondary: Machine Learning (cs.LG)
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 introduced framework, LPT++, is a comprehensive approach to long-tailed classification that combines parameter-efficient fine-tuning (PEFT) with learnable model ensembles. It enhances frozen Vision Transformers (ViTs) through three core components: universal long-tailed adaptation modules, mixture of long-tailed experts frameworks, and three-phase training paradigms. This framework is designed for long-tailed classification tasks and achieves comparable accuracy to other methods while using only a small fraction more trainable parameters. |
| Low | GrooveSquid.com (original content) | Low Difficulty Summary LPT++ is a new way to help computers classify things correctly when some categories have many more examples than others. It uses special kinds of computer vision models, called Vision Transformers (ViTs), and adds extra features to make them better at this task. The approach has three main parts: one that adapts the model to the specific task, another that combines different models’ predictions, and a training process that helps the model learn correctly. This method can work just as well as other approaches but uses only a little more computer power. |
Keywords
» Artificial intelligence » Classification » Fine tuning » Parameter efficient
