AI Paper Atlas

Research map / Learning theory and algorithms / Combinatorial and search optimization

Pinet: Optimizing hard-constrained neural networks with orthogonal projection layers

Panagiotis D. Grontas, Antonio Terpin, Efe C. Balta, Raffaello D'Andrea, John Lygeros

ICLR 2026 · Learning theory and algorithms › Combinatorial and search optimization · approach: Constrained neural network optimization · 0 citations (OpenAlex)

Read the paperCheck your idea against 137,217 papers

AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.

Measured effects reported in the paper

What was changedWhereMeasuredEffect
replace cvxpylayers projection with Pinet (GPU)projection layer · Non-convex smallbatch inference time [s]: 2.5917 -> 0.0135Improved
replace cvxpylayers projection with Pinet (CPU)projection layer · Non-convex smallsingle inference time [s]: 0.0120 -> 0.0052Improved
use Pinet instead of cvxpylayers for trainingtraining budget 25 epochs · Non-convex smalltraining wall-clock: roughly 10 minutes -> 4 and 28 seconds on the GPU and CPUImproved
use Pinet instead of JAXopt implicit layerprojection layer · Non-convex largemedian batch inference time [s]: 19.430 -> 0.2804Improved
use Pinet instead of JAXopt implicit layerprojection layer · Non-convex largetraining time: roughly 14 hours for JAXoptImproved
use Pinet batch GPU solve instead of IPOPT solverPinet · Non-convex largemedian batch inference time [s]: 10720 -> 0.2804Improved
use Pinet instead of OSQP solverPinet · Convex largemedian batch inference time [s]: 680.59 -> 0.2800Improved
OSQP parametric mode (reuse calculations) (ablation)OSQP solver · Convex largemedian batch inference time [s]: 680.59 -> 62.022Improved
Pinet instead of DC3 on single inferencePinet · Non-convex smallmedian single inference time [s]: 0.0019 -> 0.0056Hurt
increase DC3 correction steps from 10 to 50 (ablation)DC3 · Non-convex smallmedian single inference time [s]: 0.0019 -> 0.0080Hurt
increase DC3 correction steps from 10 to 50 (ablation)DC3 · Non-convex largemedian batch inference time [s]: 0.0248 -> 0.1177Hurt
decrease DC3 soft penalty from 10.0 to 2.0 and increase correction steps from 10 to 50 (ablation)DC3 · Non-convex benchmarksRS and CVNo clear effect

Closest papers by meaning

Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.

Open scime draft

Get the next scime Dispatch and a monthly field update

Free. We send a confirmation link first; every email has a one-click unsubscribe. Privacy