GridTorch released

Pytorch neural network model with grid-like activations is now open source code here

Back in 2018, Banino et al released their paper vector-based navigation using grid-like representations in artificial agents in Nature. This is one of the first instances of artificial neural networks reproducing single neuron-like statistics through training by SGD only (see also: Cueva & Wei, 2018). The authors released the code of this paper in Tensorflow here.

With the growing popularity of PyTorch, and the time crunch of my Msc., I decided to reimplement the model in PyTorch. This makes it extremely easy to instantiate different versions of the model for rapid experimentation. For even more ease-of-use I have included the weights of the trained (in Tensorflow) model, so no need to spend hours training anymore. The code of the model, training routine and evaluation can be found here.

Feel free to open a PR, or DM me (@LucasPompe) with your project (idea) -L