SOC-ICNNs admit exact dual-variable recovery of first-order geometry and local Hessians as value functions of SOCPs.
Learning Parametric Convex Functions,
5 Pith papers cite this work. Polarity classification is still indexing.
5
Pith papers citing it
years
2026 5verdicts
UNVERDICTED 5representative citing papers
SOC-ICNN generalizes ReLU-based ICNNs to SOCP, strictly expanding the class of representable convex functions while preserving similar forward-pass complexity.
An active-learning method fits nonlinear surrogates by minimizing maximum approximation error and derives worst-case error bounds over the domain.
A surrogate for parametric nonconvex optimization is constructed as the minimum of convex-monotonic function compositions and solved via parallel convex optimization, with a proof-of-concept on path tracking.
A residual-corrector neural network approximates NMPC optimal moves as state-dependent QPs for input-affine nonlinear systems, trained offline with hybrid imitation and KKT losses and validated on a robotic arm.
citing papers explorer
-
Exact Dual Geometry of SOC-ICNN Value Functions
SOC-ICNNs admit exact dual-variable recovery of first-order geometry and local Hessians as value functions of SOCPs.
-
SOC-ICNN: From Polyhedral to Conic Geometry for Learning Convex Surrogate Functions
SOC-ICNN generalizes ReLU-based ICNNs to SOCP, strictly expanding the class of representable convex functions while preserving similar forward-pass complexity.
-
Worst-case Nonlinear Regression with Error Bounds
An active-learning method fits nonlinear surrogates by minimizing maximum approximation error and derives worst-case error bounds over the domain.
-
Parametric Nonconvex Optimization via Convex Surrogates
A surrogate for parametric nonconvex optimization is constructed as the minimum of convex-monotonic function compositions and solved via parallel convex optimization, with a proof-of-concept on path tracking.
-
Amortized Nonlinear Model Predictive Control
A residual-corrector neural network approximates NMPC optimal moves as state-dependent QPs for input-affine nonlinear systems, trained offline with hybrid imitation and KKT losses and validated on a robotic arm.