NAS-Bench-Suite-Zero: Accelerating Research on Zero Cost Proxies
This work accelerates research on ZC proxies for NAS, providing a unified benchmark to improve efficiency in architecture search, though it is incremental in building on existing proxy techniques.
The paper tackles the under-studied aspects of zero-cost proxies (ZC proxies) for neural architecture search (NAS) by creating NAS-Bench-Suite, a large dataset and codebase evaluating 13 ZC proxies across 28 tasks, which enables faster experiments and reduces implementation confounds. It shows that ZC proxies capture complementary information, and by reducing biases and incorporating all proxies into NAS surrogate models, predictive performance improves by up to 42%.
Zero-cost proxies (ZC proxies) are a recent architecture performance prediction technique aiming to significantly speed up algorithms for neural architecture search (NAS). Recent work has shown that these techniques show great promise, but certain aspects, such as evaluating and exploiting their complementary strengths, are under-studied. In this work, we create NAS-Bench-Suite: we evaluate 13 ZC proxies across 28 tasks, creating by far the largest dataset (and unified codebase) for ZC proxies, enabling orders-of-magnitude faster experiments on ZC proxies, while avoiding confounding factors stemming from different implementations. To demonstrate the usefulness of NAS-Bench-Suite, we run a large-scale analysis of ZC proxies, including a bias analysis, and the first information-theoretic analysis which concludes that ZC proxies capture substantial complementary information. Motivated by these findings, we present a procedure to improve the performance of ZC proxies by reducing biases such as cell size, and we also show that incorporating all 13 ZC proxies into the surrogate models used by NAS algorithms can improve their predictive performance by up to 42%. Our code and datasets are available at https://github.com/automl/naslib/tree/zerocost.