Summary of Fedshift: Robust Federated Learning Aggregation Scheme in Resource Constrained Environment Via Weight Shifting, by Jungwon Seo et al.
FedShift: Robust Federated Learning Aggregation Scheme in Resource Constrained Environment via Weight Shifting
by Jungwon Seo, Minhoe Kim, Chunming Rong
First submitted to arxiv on: 2 Feb 2024
Categories
- Main: Machine Learning (cs.LG)
- Secondary: None
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 A novel Federated Learning (FL) approach called FedShift is introduced in this paper to address performance degradation in heterogeneous settings with mixed quantization levels. The current FL paradigm relies on a central server and suffers from significant communication overhead, impacting overall training efficiency. To mitigate this, compression techniques like quantization have been explored, but these may lead to client drift and performance degradation when different clients employ different quantization levels. FedShift employs a statistical matching mechanism based on weight shifting to align mixed-precision models, reducing model divergence and addressing quantization-induced bias. This approach can be added to existing FL optimization algorithms, enhancing their robustness and improving convergence. Experimental results show that FedShift effectively mitigates the negative impact of mixed-precision aggregation, yielding superior performance across various FL benchmarks. |
| Low | GrooveSquid.com (original content) | Low Difficulty Summary Federated Learning (FL) is a way for different devices or “clients” to work together to train a shared model without sharing their data. But this can be slow and inefficient because all the clients send their model updates back to the central server. A new approach called FedShift tries to fix this problem by letting clients use different levels of precision (like how many numbers they use) when sending their updates. This makes it harder for the central server to combine the updates, but FedShift helps by making sure all the models are aligned and similar. This makes FL work better in situations where devices have different hardware or internet connections. |
Keywords
* Artificial intelligence * Federated learning * Optimization * Precision * Quantization
