CR-DARTS: Channel Redistribution-based Differentiable Architecture Search

Ali Hassan; Tingting Zhang; Karen Egiazarian; Mårten Sjöström

doi:10.1109/ACCESS.2025.3637375

CR-DARTS: Channel Redistribution-based Differentiable Architecture Search

Ali Hassan
, Tingting Zhang
, Karen Egiazarian
, Mårten Sjöström

Computing Sciences

Research output: Contribution to journal › Article › Scientific › peer-review

2 Citations (Scopus)

2 Downloads (Pure)

Abstract

Differentiable Architecture Search (DARTS) has shown promising results in automating the design of deep learning models. However, its search process is computationally expensive because it evaluates all candidate operations simultaneously, often leading to an over-parameterized and inefficient search network. To reduce the computational cost, DARTS employs a smaller search network than the final evaluation network, which introduces an architecture optimization gap that limits real-world performance. To overcome this limitation, we introduce CR-DARTS, a multi-stage search framework designed to bridge the architecture optimization gap through an adaptive channel redistribution strategy. CR-DARTS reduces the computational complexity of the search network by compressing the shared input features among candidate operations and restoring the network dimensions via channel-wise feature concatenation. In addition, it progressively eliminates underperforming operations and redistributes the number of channels for more relevant feature extraction, thereby narrowing the gap between the search and evaluation networks.
We validated CR-DARTS on two diverse computer vision tasks to assess its generalizability. Experimental results show that the proposed search framework reduces the memory requirement of the DARTS algorithm by up to 4.3×, while addressing the architecture optimization gap. Moreover, in the evaluation phase, the discovered architecture achieves up to 25.3% reductions in computational complexity and 50.6% faster inference time compared to state-of-the-art methods, while maintaining comparable accuracy. It also produces a competitive fire segmentation network that outperforms the state-of-the-art methods while maintaining similar computational efficiency. These results demonstrate that CR-DARTS is a practical solution for neural architecture search. Source code will be made publicly available at https://github.com/Realistic3D-MIUN/CR-DARTS.

Original language	English
Pages (from-to)	201166-201182
Journal	IEEE Access
Volume	13
DOIs	https://doi.org/10.1109/ACCESS.2025.3637375
Publication status	Published - Nov 2025
Publication type	A1 Journal article-refereed

Funding

This work was supported by the European Joint Doctoral Programme on Plenoptic Imaging (PLENOPTIMA) through the European Union’s Horizon 2020 research and innovation programme under Marie Skłodowska-Curie Grant Agreement No. 956770. This research was also supported by the European Union through the EU Interreg Aurora project IMMERSE (20366448) and by Mid Sweden University internal funding. Computation and data handling were enabled by resources provided by the National Academic Infrastructure for Supercomputing in Sweden (NAISS) at the Chalmers Centre for Computational Science and Engineering (C3SE), partially funded by the Swedish Research Council through grant agreement no. 2022-06725. Additionally, the authors acknowledge the use of OpenAI’s ChatGPT for grammar and style enhancement in certain sections of this manuscript. The AI system was used to refine the clarity and fluency of the text while ensuring technical accuracy, with all content verified and approved by the authors.

Keywords

Differentiable Architecture Search
Fire Segmentation
Image Classification
Model Optimization
Neural Architecture Search

Publication forum classification

Publication forum level 1

Access to Document

10.1109/ACCESS.2025.3637375Licence: CC BY

CR-DARTS_Channel_Redistribution-Based_Differentiable_Architecture_SearchFinal published version, 2 MBLicence: CC BY

https://urn.fi/URN:NBN:fi:tuni-2025121611778Licence: CC BY

Cite this

@article{72a5f3efe11b4ac69498222567263953,

title = "CR-DARTS: Channel Redistribution-based Differentiable Architecture Search",

abstract = "Differentiable Architecture Search (DARTS) has shown promising results in automating the design of deep learning models. However, its search process is computationally expensive because it evaluates all candidate operations simultaneously, often leading to an over-parameterized and inefficient search network. To reduce the computational cost, DARTS employs a smaller search network than the final evaluation network, which introduces an architecture optimization gap that limits real-world performance. To overcome this limitation, we introduce CR-DARTS, a multi-stage search framework designed to bridge the architecture optimization gap through an adaptive channel redistribution strategy. CR-DARTS reduces the computational complexity of the search network by compressing the shared input features among candidate operations and restoring the network dimensions via channel-wise feature concatenation. In addition, it progressively eliminates underperforming operations and redistributes the number of channels for more relevant feature extraction, thereby narrowing the gap between the search and evaluation networks. We validated CR-DARTS on two diverse computer vision tasks to assess its generalizability. Experimental results show that the proposed search framework reduces the memory requirement of the DARTS algorithm by up to 4.3{\texttimes}, while addressing the architecture optimization gap. Moreover, in the evaluation phase, the discovered architecture achieves up to 25.3\% reductions in computational complexity and 50.6\% faster inference time compared to state-of-the-art methods, while maintaining comparable accuracy. It also produces a competitive fire segmentation network that outperforms the state-of-the-art methods while maintaining similar computational efficiency. These results demonstrate that CR-DARTS is a practical solution for neural architecture search. Source code will be made publicly available at https://github.com/Realistic3D-MIUN/CR-DARTS.",

keywords = "Differentiable Architecture Search, Fire Segmentation, Image Classification, Model Optimization, Neural Architecture Search",

author = "Ali Hassan and Tingting Zhang and Karen Egiazarian and M{\aa}rten Sj{\"o}str{\"o}m",

year = "2025",

month = nov,

doi = "10.1109/ACCESS.2025.3637375",

language = "English",

volume = "13",

pages = "201166--201182",

journal = "IEEE Access",

issn = "2169-3536",

publisher = "Institute of Electrical and Electronics Engineers Inc.",

}

TY - JOUR

T1 - CR-DARTS: Channel Redistribution-based Differentiable Architecture Search

AU - Hassan, Ali

AU - Zhang, Tingting

AU - Egiazarian, Karen

AU - Sjöström, Mårten

PY - 2025/11

Y1 - 2025/11

N2 - Differentiable Architecture Search (DARTS) has shown promising results in automating the design of deep learning models. However, its search process is computationally expensive because it evaluates all candidate operations simultaneously, often leading to an over-parameterized and inefficient search network. To reduce the computational cost, DARTS employs a smaller search network than the final evaluation network, which introduces an architecture optimization gap that limits real-world performance. To overcome this limitation, we introduce CR-DARTS, a multi-stage search framework designed to bridge the architecture optimization gap through an adaptive channel redistribution strategy. CR-DARTS reduces the computational complexity of the search network by compressing the shared input features among candidate operations and restoring the network dimensions via channel-wise feature concatenation. In addition, it progressively eliminates underperforming operations and redistributes the number of channels for more relevant feature extraction, thereby narrowing the gap between the search and evaluation networks. We validated CR-DARTS on two diverse computer vision tasks to assess its generalizability. Experimental results show that the proposed search framework reduces the memory requirement of the DARTS algorithm by up to 4.3×, while addressing the architecture optimization gap. Moreover, in the evaluation phase, the discovered architecture achieves up to 25.3% reductions in computational complexity and 50.6% faster inference time compared to state-of-the-art methods, while maintaining comparable accuracy. It also produces a competitive fire segmentation network that outperforms the state-of-the-art methods while maintaining similar computational efficiency. These results demonstrate that CR-DARTS is a practical solution for neural architecture search. Source code will be made publicly available at https://github.com/Realistic3D-MIUN/CR-DARTS.

AB - Differentiable Architecture Search (DARTS) has shown promising results in automating the design of deep learning models. However, its search process is computationally expensive because it evaluates all candidate operations simultaneously, often leading to an over-parameterized and inefficient search network. To reduce the computational cost, DARTS employs a smaller search network than the final evaluation network, which introduces an architecture optimization gap that limits real-world performance. To overcome this limitation, we introduce CR-DARTS, a multi-stage search framework designed to bridge the architecture optimization gap through an adaptive channel redistribution strategy. CR-DARTS reduces the computational complexity of the search network by compressing the shared input features among candidate operations and restoring the network dimensions via channel-wise feature concatenation. In addition, it progressively eliminates underperforming operations and redistributes the number of channels for more relevant feature extraction, thereby narrowing the gap between the search and evaluation networks. We validated CR-DARTS on two diverse computer vision tasks to assess its generalizability. Experimental results show that the proposed search framework reduces the memory requirement of the DARTS algorithm by up to 4.3×, while addressing the architecture optimization gap. Moreover, in the evaluation phase, the discovered architecture achieves up to 25.3% reductions in computational complexity and 50.6% faster inference time compared to state-of-the-art methods, while maintaining comparable accuracy. It also produces a competitive fire segmentation network that outperforms the state-of-the-art methods while maintaining similar computational efficiency. These results demonstrate that CR-DARTS is a practical solution for neural architecture search. Source code will be made publicly available at https://github.com/Realistic3D-MIUN/CR-DARTS.

KW - Differentiable Architecture Search

KW - Fire Segmentation

KW - Image Classification

KW - Model Optimization

KW - Neural Architecture Search

U2 - 10.1109/ACCESS.2025.3637375

DO - 10.1109/ACCESS.2025.3637375

M3 - Article

SN - 2169-3536

VL - 13

SP - 201166

EP - 201182

JO - IEEE Access

JF - IEEE Access

ER -

CR-DARTS: Channel Redistribution-based Differentiable Architecture Search

Abstract

Funding

Keywords

Publication forum classification

Access to Document

Fingerprint

Cite this