Closed zzyo-yu closed 1 year ago
Hello, and thank you for your intereset in this work.
I think you have correctly understood the code. The original implementation is in TensorFlow. When implementing the pyTorch version recently, we want to emphasize the proposed denoising self-supervised learning method, thus the sophiscated hypergraph transformer designs were omitted. If you are interested in the hypergraph transformer, you may also refer to the TensorFlow code.
Your work is great, but when I was studying your code, I didn't find the attention calculation operation in equation (3) of the paper in the torch version, maybe I overlooked the corresponding code. I wish I could get your help, thank you very much!