Jayesh K. Gupta
I am a Researcher at Microsoft Autonomous Systems where I work on improving simulations with data-driven methods.
While Julia is great, there are still a lot of existing useful differentiable Python code in PyTorch, Jax, etc. Given PyCall.jl is already so great and seamless, one might wonder what it takes to differentiate through those calls to Python functions. PyCallChainRules.jl aims for that ideal. DLPack.jl is leveraged to pass CPU or GPU arrays without any copy between Julia and Python.