Learning Private Neural Language Modeling with Attentive Aggregation
This work addresses privacy concerns in mobile keyboard suggestions for users by enhancing federated learning, though it is incremental as it builds on existing FL frameworks.
The paper tackles the problem of improving federated learning for private neural language modeling by addressing suboptimal model aggregation and lack of server-side optimization, resulting in reduced perplexity and communication costs compared to existing methods.
Mobile keyboard suggestion is typically regarded as a word-level language modeling problem. Centralized machine learning technique requires massive user data collected to train on, which may impose privacy concerns for sensitive personal typing data of users. Federated learning (FL) provides a promising approach to learning private language modeling for intelligent personalized keyboard suggestion by training models in distributed clients rather than training in a central server. To obtain a global model for prediction, existing FL algorithms simply average the client models and ignore the importance of each client during model aggregation. Furthermore, there is no optimization for learning a well-generalized global model on the central server. To solve these problems, we propose a novel model aggregation with the attention mechanism considering the contribution of clients models to the global model, together with an optimization technique during server aggregation. Our proposed attentive aggregation method minimizes the weighted distance between the server model and client models through iterative parameters updating while attends the distance between the server model and client models. Through experiments on two popular language modeling datasets and a social media dataset, our proposed method outperforms its counterparts in terms of perplexity and communication cost in most settings of comparison.