gorgonia

Why we included this project

Go teams that want to train and run neural networks without leaving their usual stack will find Gorgonia a natural fit. It behaves like a low-level Theano: you write and evaluate mathematical expressions over multidimensional tensors, and the library handles automatic and symbolic differentiation plus gradient-descent optimization. Convenience helpers cover common neural-network assembly, and CUDA-backed GPU acceleration is there for heavier workloads. Since everything compiles into an ordinary Go binary, it slots into production systems that already deploy Go services.