Sorry, it is currently not possible to directly use it for forecasting or anomaly detection. In this work, we regard the entire time series as an instance, while not focusing on timestamp-level representations. Therefore, you can use MHCCL to obtain instance-level representations. But if you wanna change the downstream task , you'd better modify the encoding method and the loss function.
Sorry, it is currently not possible to directly use it for forecasting or anomaly detection. In this work, we regard the entire time series as an instance, while not focusing on timestamp-level representations. Therefore, you can use MHCCL to obtain instance-level representations. But if you wanna change the downstream task , you'd better modify the encoding method and the loss function.