Tag
The paper introduces the Bayesian Filtering Transformer (BFT), which incorporates uncertainty into Transformers via precision-weighted attention and Kalman update residuals, improving performance on sequential recommendation and noisy LLM fine-tuning.