Summary of Data-free Federated Class Incremental Learning with Diffusion-based Generative Memory, by Naibo Wang et al.
Data-Free Federated Class Incremental Learning with Diffusion-Based Generative Memory
by Naibo Wang, Yuchen Deng, Wenjie Feng, Jianwei Yin, See-Kiong Ng
First submitted to arxiv on: 22 May 2024
Categories
- Main: Computer Vision and Pattern Recognition (cs.CV)
- Secondary: Distributed, Parallel, and Cluster Computing (cs.DC); 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 Medium Difficulty summary: Federated Class Incremental Learning (FCIL) is a crucial but underexplored challenge in federated learning (FL), involving the dynamic incorporation of new classes. Existing methods often rely on generative adversarial networks (GANs) to address privacy concerns, but GANs exhibit instability and high sensitivity, compromising their effectiveness. This paper introduces DFedDGM, a novel data-free FCIL framework that generates stable images through diffusion models. A balanced sampler and entropy-based sample filtering technique are designed to alleviate the non-IID problem in FL. Knowledge distillation is integrated with feature-based regularization for better knowledge transfer. Our framework does not incur additional communication costs compared to FedAvg. Extensive experiments on multiple datasets demonstrate significant improvements over existing baselines, such as a 4% improvement in average accuracy on Tiny-ImageNet. |
| Low | GrooveSquid.com (original content) | Low Difficulty Summary Low Difficulty summary: This paper solves a big problem called Federated Class Incremental Learning. It’s like trying to add new classes to a school without mixing up the old ones. Current methods try to use special computer vision tools to help, but they don’t work well because they’re unstable. The researchers came up with a new way to solve this problem using something called diffusion models. They also created a special tool to make sure the new information is accurate and another tool to make sure the old information stays good too. The best part is that their method doesn’t require more communication than usual. In tests, their method worked really well, improving accuracy by 4% on one dataset. |
Keywords
» Artificial intelligence » Diffusion » Federated learning » Knowledge distillation » Regularization
