gorgonia/gorgonia
Gorgonia brings differentiable computation graphs to Go: expressions over multidimensional arrays are compiled into an executable graph, differentiated automatically or symbolically, and optimized with gradient descent. It offers numerical stabilization and helpers for building neural networks, and targets performance close to Theano and TensorFlow while keeping Go's deployment model. CUDA computation is supported, and the library stays low-level, leaving model design to the developer.
Maintenance activity