AI Paper Atlas

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

Unsupervised Neural Langevin Sampler for Mixed Integer Linear Programming

Yixin Huang, Shengyu Feng, Yiming Yang

ICML 2026 · Learning theory and algorithms › Combinatorial and search optimization · approach: Predict-then-optimize learning

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
add the second-order correction term to the Erdos loss (ablation)Ours (Erdos) · CFLP-(100, 100)objective: 20983.15 -> 19117.49Improved
add the second-order correction term to the Erdos loss (ablation)Ours (Erdos) · CFLP-(200, 200)objective: 39938.12 -> 37998.29Improved
use correction term instead of an L2-style regularizer (ablation)Ours (Erdos) · CFLP-(100, 100)objective: 20217.15 -> 19117.49Improved
use correction term instead of an L2-style regularizer (ablation)Ours (Erdos) · CFLP-(200, 200)objective: 39029.12 -> 37998.29Improved
add L2-style regularizer E[x'](1-E[x']) without correction term (ablation)Ours (Erdos) · CFLP-(100, 100)objective: 20983.15 -> 20217.15Improved
train with Erdos loss instead of REINFORCE loss (ablation)Ours · CFLP-(100, 100)gap to Gurobi (%): 24.13% -> 6.24%Improved
train with Erdos loss instead of Gumbel loss (ablation)Ours · CFLP-(200, 200)gap to Gurobi (%): 18.74% -> 11.96%Improved
train with Gumbel loss instead of Erdos loss (ablation)Ours · IP (Neural Diving 400s)objective: 13.89 -> 13.82Improved
train with Gumbel loss instead of Erdos loss (ablation)Ours · LB (Neural Diving 400s)objective: 711.92 -> 706.10Improved
train with Gumbel loss instead of Erdos loss (ablation)Ours · IP (no downstream solver)objective: 483.14 -> 481.39Improved
use unsupervised Ours (Erdos) instead of supervised SL with Neural DivingOurs (Erdos) · IP (Neural Diving 400s)objective: 14.62 -> 13.89Improved
use unsupervised Ours (Erdos) instead of supervised SL on CFLPOurs (Erdos) · CFLP-(100, 100)objective: 18731.74 -> 18992.49Hurt

Earlier methods this paper compares with

Named baselines from earlier accepted papers in this paper's results tables: better than 1 of 1 on most shared results.

Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.

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