ProbXiv
Explorersign in
Problem archiveProblem record

Statement

Let N\mathcal{N} be a feedforward neural network with inputs labeled by x1,…,xnx_1,\dots,x_n, and let Fdact(N)\mathbf{F}^{\rm act}_d(\mathcal{N}) and Fd(N)\mathbf{F}_d(\mathcal{N}) denote the polynomial functional spaces associated with the network, inside Polyd,n\mathbf{Poly}_{d,n}. Polynomial PDE and PDI characterization conjecture. There exist finitely many algebraic PDEs Φj((Fxα)∣α∣≤r)=0\Phi_j\left(\left(F_{\mathbf{x}^{\mathbf{\alpha}}}\right)_{|\alpha|\leq r}\right)=0 and algebraic PDIs Θk((Fxα)∣α∣≤r)>0\Theta_k\left(\left(F_{\mathbf{x}^{\mathbf{\alpha}}}\right)_{|\alpha|\leq r}\right)>0 such that, for sufficiently large dd, Fdact(N)\mathbf{F}^{\rm act}_d(\mathcal{N}) and the corresponding semi-algebraic subset of Polyd,n\mathbf{Poly}_{d,n} share a dense Zariski open subset; the same holds for the other polynomial functional spaces Fd(N)\mathbf{F}_d(\mathcal{N}) when every neuron of N\mathcal{N} has fewer than nn inputs. This is the algebraic counterpart of the smooth characterization conjecture; the statement concerns generic polynomial functions and does not assert equality outside the shared dense Zariski open subset.

Record

Comments

Nothing has been published against this problem yet, and nothing has been said about it.

Sign in with an institutional address to take part in the discussion. Reading every thread stays open to everyone.

Sign in

Solve with an agent

Open the statement in a chat, with the problem and the ground rules already written into the prompt.

This opens a third-party site. Nothing is posted back to ProbXiv and nothing you write there is recorded here — what a model gives you is an attempt, which a person still has to check.