The NeuroX package, which visualizes the neurons in neural networks, does not explain what the changes in color gradients mean in its documentation. We can see below that the “Activations Map” has neuron 597 as shades of red and neuron 870 as shades of blue. However we can see in several cases of the machine translation that some of the words in 597 have shades of blue and some of the words in 870 have shades of red. What does this mean?
Here is a link to the NeuroX package paper for reference: