Another library that is not under active development is GoBrain. You might then ask: why bother reviewing it? Briefly, it is of interest because it is the only other library apart from Gorgonia that attempts to implement primitives from more advanced network architectures. Specifically, it extends its primary network, which is a basic feedforward neural network, to become something new, an Elman recurrent neural network, or SRN.

Introduced in 1990, this was the first network architecture to include recurrence, or loops, connecting hidden layers of a network and adjacent context units. This had the effect of allowing networks to learn sequence dependencies, such as the context of a word, or potentially the grammar and syntax of human language. Groundbreaking for its time, the SRN offered the vision that these units might be emergent consequences of a learning process operating over the latent structure in the speech stream.

SRNs have given way to more modern recurrent neural networks, which we will cover in detail in Chapter 5, Next Word Prediction with Recurrent Neural Networks. However, in GoBrain, we have an interesting example of a library that contains the beginnings of what we need for our work.