loss curve

[DingHuilsp]

The trend of the loss curve in the BYOL method is always decreasing, or does it decrease first and then increase? Because as I understand it, initially the target_network and online_network are closer, so the exported features are also closer. I looked at some blogs, and some losses do indeed decrease and then increase. However, when I train it myself, the loss keeps decreasing, reaching around 0.003, and it converges very quickly. I'm not sure if my model has collapsed.