Subhransu S. Bhattacharjee

You must create a sqlite3 socket to dump the values automatically into the database. The release will take a few more weeks as its part of a larger work, there...

Very interesting, what is your benchmark and validation performance?

My version gets around: ```Forward pass took 0.000297 - 0.0016 (Best-Worst) seconds``` while hitting ~98% accuracy on MNIST, what is the comparison of time vs accuracy for FastKAN? Edit: In...

Thanks, kindly take a look at KAL Net that I have just released if you get time.

Thanks a lot, I think it indicates that my model is a bit bulkier as expected because of the recursive stacking.

I think its a bit incomplete to not have a measure on the expressivity or performances, I think FastKAN outperforms overall?

I don't think that always holds: *KANs are RBFs*. If you approximate splines with RBFs that may be true but I do not agree with this blanket generalisation. This is...

Very interesting! Unfortunately this analysis is not novel. There is an entire literature to how there are Gaussian approximations to Neural Network outputs some which hold ar infinite width, some...

I was really hopeful 😭

Here is a comparison between the models (Same layer configurations -- for both I have used the same positional encodings):  ![W B Chart 11_5_2024,...