Research map / Architectures and efficiency / Neural architecture search
CARL: Causality-guided Architecture Representation Learning for an Interpretable Performance Predictor
Han Ji, Yuqi Feng, Jiahao Fan, Yanan Sun
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 changed | Where | Measured | Effect |
|---|---|---|---|
| add disentanglement and intervention losses (lambda1=lambda2=0.5 vs 0) (ablation) | CARL · NAS-Bench-201 (0.5% train) | Kendall's Tau: 0.601 -> 0.656 | Improved |
| add disentanglement loss only (lambda1=0.5, lambda2=0) (ablation) | CARL · NAS-Bench-201 (0.5% train) | Kendall's Tau: 0.601 -> 0.636 | Improved |
| add intervention loss only (lambda2=0.5, lambda1=0) (ablation) | CARL · NAS-Bench-201 (0.5% train) | Kendall's Tau: 0.601 -> 0.618 | Improved |
| raise lambda1 and lambda2 from 0.5 to 1 (ablation) | CARL · NAS-Bench-201 (0.5% train) | Kendall's Tau: 0.656 -> 0.617 | Hurt |
| use GCN encoder instead of GIN or GAT (ablation) | CARL · NAS-Bench-201 (0.5% train) | Kendall's Tau | Improved |
| combine node-level and edge-level disentanglement (ablation) | CARL · NAS-Bench-201 (0.5% train) | Kendall's Tau | Improved |
| use CARL instead of PINAT with 0.5% training data | CARL · NAS-Bench-201 | Kendall's Tau: 0.549 -> 0.656 | Improved |
| use CARL instead of NAR-Former with 0.5% training data | CARL · NAS-Bench-201 | Kendall's Tau: 0.590 -> 0.656 | Improved |
| use CARL with 0.02% training data vs NAR-Former with 0.04% | CARL · NAS-Bench-101 | Kendall's Tau: 0.683 vs 0.651 | Improved |
| use CARL in predictor-guided search instead of DCLP | CARL · NAS-Bench-101 search | test accuracy (%): 94.17 vs 94.17 | No clear effect |
| use CARL instead of PINAT | CARL · DARTS CIFAR-10 | accuracy (%): 97.58 -> 97.67 | Improved |
| use CARL instead of WeakNAS | CARL · TransNAS-Bench-101 Macro | average rank: 10.49 -> 5.29 | Improved |
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.
- PINAT — PINAT: A Permutation INvariance Augmented Transformer for NAS Predictor (AAAI 2023) · better on 7 of 7 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
- How Powerful are Performance Predictors in Neural Architecture Search? (NeurIPS 2021 · similarity 0.88)
- Arch-Graph: Acyclic Architecture Relation Predictor for Task-Transferable Neural Architecture Search (CVPR 2022 · similarity 0.88)
- Contrastive Neural Architecture Search With Neural Architecture Comparators (CVPR 2021 · similarity 0.87)
- Interpretable Neural Architecture Search via Bayesian Optimisation with Weisfeiler-Lehman Kernels (ICLR 2021 · similarity 0.87)
- Does Unsupervised Architecture Representation Learning Help Neural Architecture Search? (NeurIPS 2020 · similarity 0.87)
- Encodings for Prediction-based Neural Architecture Search (ICML 2024 · similarity 0.86)
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.
