reduce execution runtime

attempt to reduce the runtime of the mri model while maintaining optimal epoch count (5-10).

current runtime (Nvidia GTX 1080): 5.02 minutes @ 99.70% accuracy (3 epochs)
(high accuracy, runtime slow)

this may include:
- applying techniques such as dropout and batch norm effectively (currently implemented, may not be the best functionality)
- changing architecture

Examine the MRI model architecture in the repository, likely in a Jupyter Notebook. Look for the training loop and model definition to understand current dropout and batch norm usage. Benchmark runtime with the provided hardware spec, then experiment with architectural changes or hyperparameter tuning to reduce time while keeping accuracy above 99% over 5-10 epochs.

由索引模型根據 Issue 內容生成。