Location-Based Attention Layer

[Peidong-Wang]

It seems that attention layer type 'location' should actually be 'hybrid' [1,2].

[1] Kim, Suyoun, Takaaki Hori, and Shinji Watanabe. "Joint CTC-attention based end-to-end speech recognition using multi-task learning." Acoustics, Speech and Signal Processing (ICASSP), 2017 IEEE International Conference on. IEEE, 2017.
[2] Bahdanau, Dzmitry, et al. "End-to-end attention-based large vocabulary speech recognition." Acoustics, Speech and Signal Processing (ICASSP), 2016 IEEE International Conference on. IEEE, 2016.