Feb 12, 2020 · We propose a perturbation-based regularization - SmoothDARTS (SDARTS), to smooth the loss landscape and improve the generalizability of DARTS-based methods.

Our regularization can be easily applied to these DARTS variants and bring consistent improvements. Stabilize DARTS. After search, DARTS simply prunes out ...

Based on this observation, we propose a perturbation-based regularization - SmoothDARTS (SDARTS), to smooth the loss landscape and improve the generalizability ...

Feb 12, 2020 · This work finds that the precipitous validation loss landscape, which leads to a dramatic performance drop when distilling the final ...

Sep 12, 2024 · Based on this observation, we propose a perturbation-based regularization, named SmoothDARTS (SDARTS), to smooth the loss landscape and improve ...

2 days ago · Stabilizing differentiable architecture search via perturbation-based regularization. In International conference on machine learning, 1554 ...

One-sentence Summary: A simple, efficient and effective approach to stabilize DARTS by noise injection regarding gradient flow. Supplementary Material: zip.

In this paper, we propose Self-Distillation Differentiable Neural Architecture Search (SD-DARTS) to alleviate the discretization gap.

[Chen and Hsieh] Stabilizing Differentiable Architecture Search via Perturbation-based Regularization. ... Search via Perturbation-based Regularization. In ...

Chen, X., & Hsieh, C. -J. (2020). Stabilizing Differentiable Architecture Search via Perturbation-based Regularization.... Chen, X., Xie, ...