Limits on representing Boolean functions by linear combinations of simple functions: thresholds, ReLUs, and low-degree polynomials
share
We consider the problem of representing Boolean functions exactly by "sparse" linear combinations (over R) of functions from some "simple" class C. In particular, given C we are interested in finding low-complexity functions lacking sparse representations. When C is the set of PARITY functions or the set of conjunctions, this sort of problem has a well-understood answer, the problem becomes interesting when C is "overcomplete" and the set of functions is not linearly independent. We focus on the cases where C is the set of linear threshold functions, the set of rectified linear units (ReLUs), and the set of low-degree polynomials over a finite field, all of which are well-studied in different contexts. We provide generic tools for proving lower bounds on representations of this kind. Applying these, we give several new lower bounds for "semi-explicit" Boolean functions. For example, we show there are functions in nondeterministic quasi-polynomial time that require super-polynomial size: ∙ Depth-two neural networks with sign activation function, a special case of depth-two threshold circuit lower bounds. ∙ Depth-two neural networks with ReLU activation function. ∙ R-linear combinations of O(1)-degree F_p-polynomials, for every prime p (related to problems regarding Higher-Order "Uncertainty Principles"). We also obtain a function in E^NP requiring 2^Ω(n) linear combinations. ∙ R-linear combinations of ACC ∘ THR circuits of polynomial size (further generalizing the recent lower bounds of Murray and the author). (The above is a shortened abstract. For the full abstract, see the paper.)
READ FULL TEXT
