Two-argument activation functions learn soft XOR operations like cortical neurons
This addresses the challenge of improving neural network efficiency and robustness for AI applications, though it is incremental as it builds on existing activation function research.
The paper tackled the problem of simplifying artificial neurons compared to biologically realistic ones by learning two-argument activation functions, resulting in networks that learn faster, perform better, and are more robust than those using ReLU with matched parameters.
Neurons in the brain are complex machines with distinct functional compartments that interact nonlinearly. In contrast, neurons in artificial neural networks abstract away this complexity, typically down to a scalar activation function of a weighted sum of inputs. Here we emulate more biologically realistic neurons by learning canonical activation functions with two input arguments, analogous to basal and apical dendrites. We use a network-in-network architecture where each neuron is modeled as a multilayer perceptron with two inputs and a single output. This inner perceptron is shared by all units in the outer network. Remarkably, the resultant nonlinearities often produce soft XOR functions, consistent with recent experimental observations about interactions between inputs in human cortical neurons. When hyperparameters are optimized, networks with these nonlinearities learn faster and perform better than conventional ReLU nonlinearities with matched parameter counts, and they are more robust to natural and adversarial perturbations.