Neural networks are versatile tools for computation, having the ability to approximate a broad range of functions. An important problem in the theory of deep neural networks is expressivity; that is, we want to understand the functions that are computable by a given network. We study real, infinitely differentiable (smooth) hierarchical functions implemented by feedforward neural networks via composing simpler functions in two cases: (1) each constituent function of the composition has fewer inputs than the resulting function and (2) constituent functions are in the more specific yet prevalent form of a nonlinear univariate function (e.g., tanh) applied to a linear multivariate function. We establish that in each of these regimes, there exist nontrivial algebraic partial differential equations (PDEs) that are satisfied by the computed functions. These PDEs are purely in terms of the partial derivatives and are dependent only on the topology of the network. Conversely, we conjecture that such PDE constraints, once accompanied by appropriate nonsingularity conditions and perhaps certain inequalities involving partial derivatives, guarantee that the smooth function under consideration can be represented by the network. The conjecture is verified in numerous examples, including the case of tree architectures, which are of neuroscientific interest. Our approach is a step toward formulating an algebraic description of functional spaces associated with specific neural networks, and may provide useful new tools for constructing neural networks.

Physiological experiments have highlighted how the dendrites of biological neurons can nonlinearly process distributed synaptic inputs. However, it is unclear how aspects of a dendritic tree, such as its branched morphology or its repetition of presynaptic inputs, determine neural computation beyond this apparent nonlinearity. Here we use a simple model where the dendrite is implemented as a sequence of thresholded linear units. We manipulate the architecture of this model to investigate the impacts of binary branching constraints and repetition of synaptic inputs on neural computation. We find that models with such manipulations can perform well on machine learning tasks, such as Fashion MNIST or Extended MNIST. We find that model performance on these tasks is limited by binary tree branching and dendritic asymmetry and is improved by the repetition of synaptic inputs to different dendritic branches. These computational experiments further neuroscience theory on how different dendritic properties might determine neural computation of clearly defined tasks.

Any function can be constructed using a hierarchy of simpler functions through compositions. Such a hierarchy can be characterized by a binary rooted tree. Each node of this tree is associated with a function that takes as inputs two numbers from its children and produces one output. Since thinking about functions in terms of computation graphs is becoming popular, we may want to know which functions can be implemented on a given tree. Here, we describe a set of necessary constraints in the form of a system of nonlinear partial differential equations that must be satisfied. Moreover, we prove that these conditions are sufficient in contexts of analytic and bit-valued functions. In the latter case, we explicitly enumerate discrete functions and observe that there are relatively few. Our point of view allows us to compare different neural network architectures in regard to their function spaces. Our work connects the structure of computation graphs with the functions they can implement and has potential applications to neuroscience and computer science.