search

HerrmannM / paper-2021-ADTW

ADTW demonstration application - implemented with EAP
BSD 3-Clause "New" or "Revised" License
1 stars 0 forks source link

Differentiablity and Soft DTW #1

Open stellarpower opened 3 months ago

stellarpower commented 3 months ago

Hi,

Looks like amercing might be exactly what I am looking for in my loss function.

I was wondering if you'd ever considered integrating it with the Soft DTW algorithm before, or if you know any reason why it might not be differentiable. Looking through the paper, I assume given that the penalty is pretty straightforward, it should be possible to express it in a way that can be differentiated and add that into the calculation for the backwards pass of the soft version. But my maths isn't that great.

Have you ever looked into this at all?

Thanks!

HerrmannM commented 2 months ago

Hi,

Apologies for the delay.

Unfortunately, I changed job and I am no longer working on this subject. I'll signal the issue to my former coworker (but I can't promise anything).

All the bests, Mat

stellarpower commented 2 months ago

Thanks a lot, appreciate it!