REVIEW 2 cited by
Can stable and accurate neural networks be computed? -- On the barriers of deep learning and Smale's 18th problem
Not yet reviewed by Pith; the record is open.
This paper has not been read by Pith yet. Machine review is queued; the pith claim, tier, and objections will appear here once it completes.
SPECIMEN: schema-true, not a live event
T0 review · schema-true
One-sentence machine reading of the paper's core claim.
pith:XXXXXXXX · record.json · timestamp
Signed reviews
abstract
Deep learning (DL) has had unprecedented success and is now entering scientific computing with full force. However, current DL methods typically suffer from instability, even when universal approximation properties guarantee the existence of stable neural networks (NNs). We address this paradox by demonstrating basic well-conditioned problems in scientific computing where one can prove the existence of NNs with great approximation qualities, however, there does not exist any algorithm, even randomised, that can train (or compute) such a NN. For any positive integers $K > 2$ and $L$, there are cases where simultaneously: (a) no randomised training algorithm can compute a NN correct to $K$ digits with probability greater than $1/2$, (b) there exists a deterministic training algorithm that computes a NN with $K-1$ correct digits, but any such (even randomised) algorithm needs arbitrarily many training data, (c) there exists a deterministic training algorithm that computes a NN with $K-2$ correct digits using no more than $L$ training samples. These results imply a classification theory describing conditions under which (stable) NNs with a given accuracy can be computed by an algorithm. We begin this theory by establishing sufficient conditions for the existence of algorithms that compute stable NNs in inverse problems. We introduce Fast Iterative REstarted NETworks (FIRENETs), which we both prove and numerically verify are stable. Moreover, we prove that only $\mathcal{O}(|\log(\epsilon)|)$ layers are needed for an $\epsilon$-accurate solution to the inverse problem.
Forward citations
Cited by 2 Pith papers
-
On the computation of geometric features of spectra of linear operators on Hilbert spaces
First algorithms and solvability-complexity classifications for computing Lebesgue measure, capacity, and fractal dimensions of spectra of infinite-dimensional linear operators.
-
Computing Spectral Measures and Spectral Types
First general algorithms compute spectral measures, point/continuous/singular decompositions, functional calculus, and Radon-Nikodym derivatives for self-adjoint or unitary operators with known column decay, with Solv...
Discussion (0). Continue with ORCID to comment.