AI Paper Atlas

Research map / Architectures and efficiency / Neural architecture search

Searching for BurgerFormer with Micro-Meso-Macro Space Design

Longxing Yang, Yu Hu, Shun Lu, Zihao Sun, Jilin Mei, Yinhe Han, Xiaowei Li

ICML 2022 · Architectures and efficiency › Neural architecture search · approach: Differentiable architecture search (DARTS)

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
hybrid sampling (SPOS + sandwich-style min/max macro sampling) instead of random search (ablation)supernet training · ImageNet-1k, 1G FLOPsTop-1 acc (%): +2.4 vs random searchImproved
hybrid sampling instead of SPOS (ablation)supernet training · ImageNet-1k, 1G FLOPsTop-1 acc (%): +1.0 vs SPOSImproved
hybrid sampling instead of sandwich sampling (ablation)supernet training · ImageNet-1k, 1G FLOPsTop-1 acc (%): +0.5 vs SandwichImproved
include first Norm-Op-Norm-Act (bread) in hamburger block (ablation)BurgerFormer-Tiny · ImageNet-1kTop-1 acc (%): +0.3 (78.0 vs 77.7)Improved
include second Norm-Op-Norm-Act (bread) in hamburger block (ablation)BurgerFormer-Tiny · ImageNet-1kTop-1 acc (%): +0.8 (78.0 vs 77.2)Improved
include meat Norm-Op-Norm-Act in hamburger block (ablation)BurgerFormer-Tiny · ImageNet-1kTop-1 acc (%): +1.1 (78.0 vs 76.9)Improved
searched BurgerFormer vs manually designed and auto-searched Transformers at similar FLOPsBurgerFormer-Tiny/Base/Large · ImageNet-1kTop-1 acc (%): BurgerFormer-Base 82.7 vs Swin-T 81.3 at fewer FLOPsImproved
BurgerFormer-Large matches Swin-S accuracy at fewer FLOPsBurgerFormer-Large · ImageNet-1kTop-1 acc (%): 83.0 vs 83.0 with 25% fewer FLOPsNo clear effect
BurgerFormer-Base as backbone instead of PVT-Small/Swin-T in detectorsRetinaNet and Mask R-CNN on COCO · COCOAP: RetinaNet 41.2 vs PVT-Small 40.4 (better); Mask R-CNN box AP 44.0 vs Swin-T 43.7 (better), but Swin-T is 48.0M params vsMixed

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