We are please to release owl-symbolic 0.1.0. It fully supports defining a computation graph and running on accelerators (TPU/GPU) via ONNX specification. It also aims to support converting an Owl computation graph into symbolic representation and then to ONNX model. The module also has some cool features like converting a computation graph into LaTeX string, and then showing the result in a web UI, etc.
We implements a full neural network module atop of it (the interface of which is basically identical to that in Owl’s core). It turns out that the design of owl-symbolic
is so nice that the DNN module only has 179 LOC! You can easily define popular DNN architectures such as Inception, ResNet, VGG, etc. just like in Owl.
This is still an on-going project and a lot remains to be done. Despite its name, owl-symbolic
does not do any useful computer algebra (CAS) stuff at the moment, but CAS is indeed on our TODO.
For more information, please check out the related section in Owl tutorial book.