About SleepEEG dataset classification performance

[jaeho3690]

First and foremost, I would like to express my deep appreciation for your recent publication. Your research has been helpful for my current research, and I've found the methodology and results particularly compelling. I am reaching out with a few questions about Table 5, titled "Standard Time Series Classification Setting," as detailed in your paper.

1. Could you please clarify if the "Standard Setting" referenced in your study is based on a Supervised Classification framework, specifically utilizing cross-entropy for model training?
2. Regarding the dataset and its partitioning: did your work employ the data split provided by the TF-C repository? My attempts to replicate your findings have yielded an accuracy range of approximately 50-60%, which significantly diverges from the outcomes reported in your document.
3. Would it be possible for you to share the hyperparameters used in the SleepEEG Classification, as outlined in Table 5? Understanding these settings would be incredibly beneficial for my research.

Thank you once again for your invaluable contribution to the field.

Warm regards,

Jaeho

[seunghan96]

First of all, I apologize for the late response.

1. Here's the comparison of "standard setting" and "transfer setting":

   • Standard setting: Pretrain dataset == Fine-tune dataset
   • Transfer setting: Pretrain dataset != Fine-tune dataset

2. Sorry for the mistake. There was an error in our documentation; we used the "Epilepsy" dataset (following SimMTM) and mistakenly referred to it as "SleepEEG". Sorry for the confusion.

3. I will also add a hyperparameter for "standard classification" in the appendix!

Once again, thank you for pointing that out!

[jaeho3690]

Thank you!