Challenge:
to design an explanation mechanism that is tightly integrated into the machine learning model.

Interesting future work:

1. to discover which part(s) in the model structure explains its different functions and play a major role in the performance improvement or decline of each iteration.
2. to better illustrate the rationale behind the model and the decisions made. ***

Challenge:

1. the difficulty of selecting and comparing representative snapshots from a large number of snapshots.
2. automatically and accurately detecting anomalies in the training process.

to combine system imitative guidance and user initiative guidance as mixed initiative guidance to maximize the value of both. ???