Reproduce “Prediction with Foundation Models under Deployment Budgets” 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
Zhaohui Xu, Xiaofei Xu, Doudou Zhou
The authors introduce ACES, which combines task features with model predictions and learns which predictions to query. ACES-T uses a fixed affordable collection, while ACES-I adapts queries across inputs under an average-cost budget. On laptop listings at $4.08 per 10,000 predictions, ACES-I queries Gemini-J for 40% and achieves log-RMSE 0.340 versus 0.347 for ACES-T, at higher cost; this result is application-specific.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.