I’m working on developing a attractor network that requires the units of a layer to connect to each other, in addition to the output and input connections. I’m not quite sure how it can implemented without using a recurrent layer, or am I restricted to using a recurrent layer with some modifications? Thanks.