Share your thoughts, 1 month free Claude Pro on usSee more
WorkDL logo mark

CARL: Causality-guided Architecture Representation Learning for an Interpretable Performance Predictor

About

Performance predictors have emerged as a promising method to accelerate the evaluation stage of neural architecture search (NAS). These predictors estimate the performance of unseen architectures by learning from the correlation between a small set of trained architectures and their performance. However, most existing predictors ignore the inherent distribution shift between limited training samples and diverse test samples. Hence, they tend to learn spurious correlations as shortcuts to predictions, leading to poor generalization. To address this, we propose a Causality-guided Architecture Representation Learning (CARL) method aiming to separate critical (causal) and redundant (non-causal) features of architectures for generalizable architecture performance prediction. Specifically, we employ a substructure extractor to split the input architecture into critical and redundant substructures in the latent space. Then, we generate multiple interventional samples by pairing critical representations with diverse redundant representations to prioritize critical features. Extensive experiments on five NAS search spaces demonstrate the state-of-the-art accuracy and superior interpretability of CARL. For instance, CARL achieves 97.67% top-1 accuracy on CIFAR-10 using DARTS.

Han Ji, Yuqi Feng, Jiahao Fan, Yanan Sun• 2025

Related benchmarks

TaskDatasetResultRank
Neural Architecture SearchNAS-Bench-201 ImageNet-16-120 (test)
Accuracy46.61
140
Neural Architecture SearchCIFAR-10 NAS-Bench-201 (val)
Accuracy91.49
111
Neural Architecture SearchNAS-Bench-201 CIFAR-10 (test)
Accuracy94.31
110
Neural Architecture SearchImageNet16-120 NAS-Bench-201 (val)
Accuracy46.27
104
Neural Architecture SearchNAS-Bench-201 CIFAR-100 (test)
Accuracy72.92
103
Neural Architecture SearchCIFAR-100 NAS-Bench-201 (val)
Accuracy72.84
92
Architecture Performance PredictionNAS-Bench-101 (test)
Kendall's Tau0.68
49
Performance PredictionNAS-Bench-101 (val)
Accuracy94.71
11
Performance PredictionNAS-Bench-201 CIFAR-10 (test)
Kendall's Tau0.64
9
Performance PredictionNAS-Bench CIFAR-100 201 (test)
Kendall's Tau0.65
9
Showing 10 of 11 rows

Other info

Follow for update