Backward mode about directloss minimization

[zhangrentu]

❓ Questions and Help

While using the backward mode DLM process, it was observed that there are fluctuations during the descent of the loss function. Apart from adjusting dlm_epsilon, what other parameters can be adjusted? The example link used is [provide the example link https://github.com/facebookresearch/theseus/blob/main/examples/simple_example.py