Finite training error extends the overlap-gap threshold in binary perceptrons so wide, algorithmically reachable basins persist and still generalize where zero-error solutions are hard.
Title resolution pending
1 Pith paper cite this work, alongside 6 external citations. Polarity classification is still indexing.
1
Pith paper citing it
6
external citations · OpenAlex
fields
cond-mat.dis-nn 1years
2026 1verdicts
CONDITIONAL 1representative citing papers
citing papers explorer
-
On the robustness of noisy solutions in non-convex neural networks
Finite training error extends the overlap-gap threshold in binary perceptrons so wide, algorithmically reachable basins persist and still generalize where zero-error solutions are hard.