Neural networks with piecewise linear activation functions, such as rectified linear units (ReLU) or maxout, are among the most fundamental models in modern machine learning. We make a step towards proving lower bounds on the size of such neural networks by linking their representative capabilities to the notion of the extension complexity $\mathrm{xc}(P)$ of a polytope $P$, a well-studied quantity in combinatorial optimization and polyhedral geometry. To this end, we propose the notion of virtual extension complexity $\mathrm{vxc}(P)=\min\{\mathrm{xc}(Q)+\mathrm{xc}(R)\mid P+Q=R\}$. This generalizes $\mathrm{xc}(P)$ and describes the number of inequalities needed to represent the linear optimization problem over $P$ as a difference of two linear programs. We prove that $\mathrm{vxc}(P)$ is a lower bound on the size of a neural network that optimizes over $P$. While it remains open to derive strong lower bounds on virtual extension complexity, we show that powerful results on the ordinary extension complexity can be converted into lower bounds for monotone neural networks, that is, neural networks with only nonnegative weights. Furthermore, we show that one can efficiently optimize over a polytope $P$ using a small virtual extended formulation. We therefore believe that virtual extension complexity deserves to be studied independently from neural networks, just like the ordinary extension complexity. As a first step in this direction, we derive an example showing that extension complexity can go down under Minkowski sum.

The exact nonnegative matrix factorization (exact NMF) problem is the following: given an $m$-by-$n$ nonnegative matrix $X$ and a factorization rank $r$, find, if possible, an $m$-by-$r$ nonnegative matrix $W$ and an $r$-by-$n$ nonnegative matrix $H$ such that $X = WH$. In this paper, we propose two heuristics for exact NMF, one inspired from simulated annealing and the other from the greedy randomized adaptive search procedure. We show that these two heuristics are able to compute exact nonnegative factorizations for several classes of nonnegative matrices (namely, linear Euclidean distance matrices, slack matrices, unique-disjointness matrices, and randomly generated matrices) and as such demonstrate their superiority over standard multi-start strategies. We also consider a hybridization between these two heuristics that allows us to combine the advantages of both methods. Finally, we discuss the use of these heuristics to gain insight on the behavior of the nonnegative rank, i.e., the minimum factorization rank such that an exact NMF exists. In particular, we disprove a conjecture on the nonnegative rank of a Kronecker product, propose a new upper bound on the extension complexity of generic $n$-gons and conjecture the exact value of (i) the extension complexity of regular $n$-gons and (ii) the nonnegative rank of a submatrix of the slack matrix of the correlation polytope.