This is a page that curates AI-related papers published worldwide. All content here is summarized using Google Gemini and operated on a non-profit basis. Copyright for each paper belongs to the authors and their institutions; please make sure to credit the source when sharing.
This paper highlights that while existing federated learning approaches have focused on data privacy, the emergence of large-scale language models (LLMs) has heightened the importance of intellectual property (IP) protection. Therefore, a novel federated learning approach capable of protecting both sensitive data and proprietary models is needed. To address this, we propose a novel federated learning method, FedQSN. FedQSN randomly masks some model parameters and quantizes the remaining parameters, enabling the model transmitted by the server to the client to act as a privacy-preserving proxy. Experimental results across various models and tasks demonstrate that FedQSN enhances model parameter protection compared to existing methods while maintaining robust model performance in a federated learning environment.
Takeaways, Limitations
•
Takeaways:
◦
We present a novel approach to intellectual property protection in federated learning of large-scale language models.
◦
We present an effective method to enhance the privacy protection of model parameters.
◦
We experimentally demonstrate that privacy can be improved without compromising the performance of existing federated learning.
•
Limitations:
◦
There is a lack of theoretical analysis on the safety of the proposed method.
◦
Further evaluation of resistance to various attack scenarios is needed.
◦
The experimental results are limited to specific models and tasks and require further research to determine their generalizability.
