Think about a step to understand where errors are made, maybe after squeeze

[SebChw]

No description provided.

[kordc]

@SebChw could you elaborate more on this? I don't think I understand this issue

[SebChw]

Yes:

I imagine it like this we are still unhappy with our results and we are thinking on what causes bad results.

Do we always mess up between 2 classes? Are some particular examples harder?

[trebacz626]

Shal we add this to mnist tutorail and craete confusion matrix and save examples with highest loss?