void brook
I've drafted a proposal for a Java CNN Builder for an open-source Java ML library and would greatly appreciate your feedback. The main goal is to provide an intuitive and easy way to create CNNs. Feel free to respond directly on GitHub. Thank you all! https://github.com/Samyssmile/edux/discussions/50
analog nestBOT
⌛ This post has been reserved for your question.
Hey @void brook! Please use
/closeor theClose Postbutton above when your problem is solved. Please remember to follow the help guidelines. This post will be automatically closed after 300 minutes of inactivity.
TIP: Narrow down your issue to simple and precise questions to maximize the chance that others will reply in here.
digital harness
Why make it limited to CNNs?
Also, is it possible to compose (already bult) architectures?
digital harness
Why make it limited to CNNs?
or what makes this specifically a CNN builder as opposed to an NN builder?
Also why?
What's the point of using virtual threads for CPU bound stuff?
I can tell you: nothing
virtual threads aren't magically faster threads
void brook
No they not, but for our Benchmark we wanna have them. We're running benchmarks on 4000x4000 matrices to explore practical impacts. We also use Strassen Approach, Threaded, CUDA, Single Threads classic.
Btw. CUDA is magicly instant 
void brook
or what makes this specifically a CNN builder as opposed to an NN builder?
Nothing, our MLP build approach will be replaced by this one.
digital harness
Nothing, our MLP build approach will be replaced by this one.
Then why are you calling it CNN builder?
Does it even support 3D convolutions?
void brook
Does it even support 3D convolutions?
point cloud classification firmly is planned on development roadmap