Reproduce “AHEAD: Adaptive Neural Acceleration for Large-Scale Linear Programming” on OpenResearch
Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.
Universal app for Apple silicon and Intel. Requires macOS 11 or later.
About this paper
Sihao Liu, Zhichao Hou, Huayue Gu, Raiyan Rahman, Liang Wu, Ruozhou Yu, Xiaorui Liu
AHEAD steers large linear programs by rescaling a first-order solver’s primal and dual step sizes rather than predicting solutions outright. A shared controller uses solver statistics, while a separate learned policy selects when to hand control back. On 20 held-out, size-shifted synthetic packing problems, it reports 11.59× aggregate end-to-end speedup at tolerance 10⁻⁴ versus cold-start Gurobi GPU-PDHG; evidence is limited to synthetic packing and PageRank LPs.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.