AI Paper Atlas

Research map / Architectures and efficiency / RNNs and normalization

HALO: Hardware-Aware Learning to Optimize

Chaojian Li, Tianlong Chen, Haoran You, Zhangyang Wang, Yingyan Lin

ECCV 2020 · Architectures and efficiency › RNNs and normalization · approach: Distributed LLM inference systems · 16 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 Jacobian regularizer to vanilla HALO (ablation)HALO · CIFAR-10-A wider optimizeeadaptation accuracy after convergenceImproved
add structural sparsity regularizer to HALO with Jacobian regularizer (ablation)HALO · CIFAR-10-A wider optimizeeaccuracy under same energy: up to 11.70% higher accuracyImproved
add structural sparsity regularizer to HALO with Jacobian regularizer (ablation)HALO · CIFAR-10-A wider optimizeetraining loss under same energy: up to 16.72% lowerImproved
add both regularizers to vanilla HALO to beat Adam (ablation)HALO · CIFAR-10-A wider optimizeeloss and accuracy after convergenceImproved
add structural sparsity regularizer to Adam (ablation)Adam · CIFAR-10-A wider optimizeeadaptation accuracyHurt
use progressively increased update probability 10%-30%-50% instead of 50%-30%-10% (ablation)HALO · CIFAR-10-A wider optimizeeaccuracy under same energy: up to 20.28% AccImproved
use progressively increased update probability instead of uniform 30%-30%-30% (ablation)HALO · CIFAR-10-A wider optimizeeaccuracy under same energyImproved
use HALO instead of Hierarchical-L2OHALO · TDPfinal test accuracy / energy: 80.28 vs 79.19; 31.8% less energyImproved
use HALO instead of Hierarchical-L2OHALO · GSADfinal test accuracy / energy: 87.00 vs 80.31; 58.6% - 60.0% less energyImproved
use HALO instead of Hierarchical-L2OHALO · SPfinal test accuracy / energy: 98.16 vs 97.20; 44.3% - 45.1% less energyImproved
use HALO instead of AdamHALO · TDPenergy at 3k iterations: 0.13 -> 0.45 kJHurt

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