Fovea Transformer: Efficient Long-Context Modeling with Structured Fine-To-Coarse Attention | IEEE Conference Publication | IEEE Xplore