A typical biological neuron, such as a pyramidal neuron of the neocortex, receives thousands of afferent synaptic inputs on its dendrite tree and sends the efferent axonal output downstream. In typical artificial neural networks, dendrite trees are modeled as linear structures that funnel weighted synaptic inputs to the cell bodies. However, numerous experimental and theoretical studies have shown that dendritic arbors are far more than simple linear accumulators. That is, synaptic inputs can actively modulate their neighboring synaptic activities; therefore, the dendritic structures are highly nonlinear. In this study, we model such local nonlinearity of dendritic trees with our dendritic neural network (DENN) structure and apply this structure to typical machine learning tasks. Equipped with localized nonlinearities, DENNs can attain greater model expressivity than regular neural networks while maintaining efficient network inference. Such strength is evidenced by the increased fitting power when we train DENNs with supervised machine learning tasks. We also empirically show that the locality structure can improve the generalization performance of DENNs, as exemplified by DENNs outranking naive deep neural network architectures when tested on 121 classification tasks from the UCI machine learning repository.

This paper presents D-Nets, an architecture loosely inspired by the dendrites of biological neurons. In a D-Net, each neuron receives input from the previous layer as the maxpool of linear combinations of disjoint random subsets of that layer's outputs. The authors show that this approach outperforms self-normalizing neural networks and other advanced approaches on the UCI collection of datasets (as well as outperforming simple non-convolutional approaches to MNIST and CIFAR). They provide an intuition that greater fan-in to non-linearities leads to a greater number of linear regions and thus, perhaps, greater expressibility. I am still quite surprised that such a simple method performs so well, but the experimental setup seems sound. For example, how does the optimal number of branches grow with the size of the layer?

Despite the remarkable success of Graph Neural Networks (GNNs), the common belief is that their representation power is limited and that they are at most as expressive as the Weisfeiler-Lehman (WL) algorithm. In this paper, we argue the opposite and show that standard GNNs, with anonymous inputs, produce more discriminative representations than the WL algorithm. Our novel analysis employs linear algebraic tools and characterizes the representation power of GNNs with respect to the eigenvalue decomposition of the graph operators. We prove that GNNs are able to generate distinctive outputs from white uninformative inputs, for, at least, all graphs that have different eigenvalues. We also show that simple convolutional architectures with white inputs, produce equivariant features that count the closed paths in the graph and are provably more expressive than the WL representations. Thorough experimental analysis on graph isomorphism and graph classification datasets corroborates our theoretical results and demonstrates the effectiveness of the proposed approach.