Fig. 5: Activity comparison of an OReN and a saccadic network using spiking neurons.
From: On the visual analytic intelligence of neural networks
N = 32. a OReN identifies the oddity through calculation of gθ for pairs of input embeddings, followed by aggregation and scoring through fφ. The embedding pairs comprising the oddity as the first element lead to a visually distinctive gθ activation map for the oddity. b The saccadic network receives a temporal stream of inputs that lead to the evolution of the internal states, Vm, visualized for Layer 3. c Membrane potentials sorted by the order of the first OReN embedding pair item exhibit similar visual distinction for the oddity. Visualized values were normalized row-wise, i.e., over 32 neurons.
