Zhanpeng Zeng

Thank you so much for the prompt response! This is helpful.

Hi Dennis, The algorithm guarantees the convergence of approximation as number of landmarks increases, and the error of approximation is bounded in terms of matrix norm. If you want the...

Hi, sorry for the late response. We have actually asked the authors of LRA about this issue, but the problem is not completely resolved. https://github.com/google-research/long-range-arena/issues/18 We suspect that the difference...

Are you using code from [LRA](https://github.com/mlpen/Nystromformer/tree/main/reorganized_code/LRA)? This [config](https://github.com/mlpen/Nystromformer/blob/main/reorganized_code/LRA/models/softmax/config.json) file is an example. To run LRA on other attentions, you can modify the "attn_type" (see possible attention methods in [code](https://github.com/mlpen/Nystromformer/blob/main/reorganized_code/encoders/backbones/efficient_attention.py)) and...