AI Paper Atlas

Research map / Architectures and efficiency / RNNs and normalization

AdaTune: Adaptive Tensor Program Compilation Made Efficient

Menghao Li, Minjia Zhang, Chi Wang, Mingqin Li

NeurIPS 2020 · Architectures and efficiency › RNNs and normalization · approach: Efficient network compression and training · 13 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
add adaptive evaluator (CV-based early termination of hardware measurement) (ablation)AdaTune · ResNet-18 on CPU, per-iteration costhardware measurement time: 2.5× reductionImproved
add adaptive evaluator on top of RFEI + CSA (ablation)AdaTune · 4 tuning workloadswall-clock search timeImproved
replace XGBoost with random forest + expected improvement surrogate (ablation)AutoTVM baseline · search effectiveness per iterationiterations to find best planMixed
replace fixed epsilon-greedy with contextual simulated annealing (ablation)AdaTune · search effectiveness per iterationsearch effectivenessMixed
use AdaTune instead of AutoTVM under same time budgetBatched GEMM (GPU) · tuningbest GFLOPS: up to 115% higher GFLOPSImproved
use AdaTune instead of AutoTVM for end-to-end optimizationResNet-18 on GPU · optimization timehours: 22.6 hours -> 9.6 hoursImproved
use AdaTune instead of AutoTVM for final inference timeResNet-18 · inference speedinference speed: up to 4.6% fasterImproved

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