Extended Data Table 1 A unified description of neural network models

From: Towards artificial general intelligence with hybrid Tianjic chip architecture

  1. The vector/matrix operations and transformations are assigned into several compartments, including axon, synapse, dendrite, soma and router. Each compartment supports the functions shown; by combining these, various neuroscience-oriented and computer-science-oriented models can be realized. In SNN: s, spike; se, external spike; Spike_fire, spike firing; Thresh_comp, threshold comparison; τ, time constant; t, time step; V, membrane potential; Vr1, rest potential; V_reset, membrane potential reset; V_update, membrane potential update. In MLP/CNN: x/y, input/output activation; b, bias; \( \circledast \), convolution. For other abbreviations and variables, please refer to Extended Data Figs. 1, 3.
Back to article page