Replacing it with a dataset that did not appear in the paper resulted in very poor results.

[Guuuli]

When I replaced the dataset, which was not present in the paper, the result was very poor. ┃ Epoch ┃ LOSS ┃ ACC ┃ MF1 ┃ KAPPA ┃ LOSS ┃ ACC ┃ MF1 ┃ KAPPA ┃ ┡━━━━━━━╇━━━━━━━━╇━━━━━━━━╇━━━━━━━━╇━━━━━━━━╇━━━━━━━━╇━━━━━━━━╇━━━━━━━━╇━━━━━━━┩ │ 1 │ 0.125 │ 0.027 │ 0.014 │ 0.005 │ 0.125 │ 0.030 │ 0.014 │ 0.010 │ │ 2 │ 0.127 │ 0.037 │ 0.008 │ 0.006 │ 0.126 │ 0.039 │ 0.012 │ 0.009 │ │ 3 │ 0.125 │ 0.028 │ 0.007 │ 0.007 │ 0.125 │ 0.042 │ 0.011 │ 0.010 │ │ 4 │ 0.126 │ 0.023 │ 0.009 │ 0.004 │ 0.125 │ 0.022 │ 0.011 │ 0.006 │ │ 5 │ 0.124 │ 0.028 │ 0.010 │ 0.005 │ 0.126 │ 0.027 │ 0.010 │ 0.004 │

[blacksnail789521]

In contrastive learning, especially when using an asymmetrical design, be cautious of model collapse (refer to the SimSiam paper for more details). Also, please provide more information about the dataset you are using.

[Guuuli]

Please check your Google email, thank you!