Extends the hyperplane arrangement framework for neural network expressivity to discriminantal arrangements, with compatible piecewise linear functions characterized by circuit relations and matroid dimension via Mobius inversion.
Title resolution pending
1 Pith paper cite this work. Polarity classification is still indexing.
1
Pith paper citing it
fields
math.CO 1years
2026 1verdicts
UNVERDICTED 1representative citing papers
citing papers explorer
-
Piecewise linear functions and neural network expressivity via discriminantal arrangements
Extends the hyperplane arrangement framework for neural network expressivity to discriminantal arrangements, with compatible piecewise linear functions characterized by circuit relations and matroid dimension via Mobius inversion.