IRO Journals
Journal of Soft Computing Paradigm

Adaptive Multi-Objective Deep Reinforcement Learning for Policy-Driven Neural Architecture Search of Resource-Efficient CNNs

Open Access
Published: 12 October, 2026
Pages: 329-340
  • Ram Teja T. , Ram Teja T.
    Student, Department of Electronics and Communication Engineering, QISCET, Ongole, India
    Student, Department of Electronics and Communication Engineering, QISCET, Ongole, India
  • Madhav G. , Madhav G.
    Student, Department of Electronics and Communication Engineering, QISCET, Ongole, India
    Student, Department of Electronics and Communication Engineering, QISCET, Ongole, India
  • Sowmya Latha P. , Sowmya Latha P.
    Student, Department of Electronics and Communication Engineering, QISCET, Ongole, India
    Student, Department of Electronics and Communication Engineering, QISCET, Ongole, India
  • Santhosh Kumar CH.V.V. , Santhosh Kumar CH.V.V.
    Trainer, Bytexl, India
    Trainer, Bytexl, India
  • Vamsi G. , Vamsi G.
    Student, Department of Electronics and Communication Engineering, QISCET, Ongole, India
    Student, Department of Electronics and Communication Engineering, QISCET, Ongole, India
  • Yogendra B. Yogendra B.
    Student, Department of Electronics and Communication Engineering, QISCET, Ongole, India
    Student, Department of Electronics and Communication Engineering, QISCET, Ongole, India
view PDF
view PDF

How to Cite

T., Ram Teja, Madhav G., Sowmya Latha P., Santhosh Kumar CH.V.V., Vamsi G., and Yogendra B. 2026. “Adaptive Multi-Objective Deep Reinforcement Learning for Policy-Driven Neural Architecture Search of Resource-Efficient CNNs”. Journal of Soft Computing Paradigm 8 (4): 329-40. https://doi.org/10.36548/jscp.2026.4.001.
ACM ACS APA ABNT Chicago Harvard IEEE MLA Turabian Vancouver AMA

Download Citation

Endnote/Zotero/Mendeley (RIS) BibTeX

Keywords

Neural Architecture Search
Deep Reinforcement Learning
Multi-Objective Optimization
Policy-Driven Architecture Generation
Adaptive Reward Optimization
Accuracy-Complexity Trade-off
Resource-Aware CNN Design

Abstract

Neural Architecture Search (NAS) automates the process of designing neural networks but the focus is mostly on prediction accuracy without considering the model complexity and computational cost. This research study proposes a new framework called Adaptive Multi-Objective Policy-Driven Neural Architecture Search (AMO-PDNAS) based on Deep Reinforcement Learning (DRL). The CNN architecture design is formulated using Markov Decision Process. A parameterized policy network generates CNN architecture parameters, which include kernel size, number of filters, activation functions, and dropout rate. Reward weighting is adapted through each search cycle in an attempt to shift from an explorative approach driven by accuracy towards an approach driven by both accuracy and efficiency considerations. Candidate architectures are learned and evaluated in cycles and feedback is provided to the policy based on the REINFORCE algorithm. Results on the MNIST problem show robust behavior of searching for high-accuracy architectures along with efficient trade-offs of accuracy and complexity.

References

  1. He, Kaiming, Xiangyu Zhang, Shaoqing Ren, and Jian Sun. ”Deep Residual Learning for Image Recognition.” In Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition, 2016, 770-778.
  2. Cai, Han, Ligeng Zhu, and Song Han. ”Proxylessnas: Direct Neural Architecture Search on Target Task and Hardware.” arXiv preprint, 2018, arXiv:1812.00332.
  3. Baker, Bowen, Otkrist Gupta, Nikhil Naik, and Ramesh Raskar. ”Designing Neural Network Architectures Using Reinforcement Learning.” arXiv preprint, 2016,arXiv: 1611.02167.
  4. Poyser, Matt, and Toby P. Breckon. ”Neural Architecture Search: A Contemporary Literature Review for Computer Vision Applications.” Pattern Recognition, 2024, vol. 147: 110052.
  5. Wang, Zhaozhi, Kefan Su, Jian Zhang, Huizhu Jia, Qixiang Ye, Xiaodong Xie, and Zongqing Lu. ”Multi-agent Automated Machine Learning.” In IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), IEEE, 2023, 11960-11969.
  6. Li, Yuhong, Jiajie Li, Cong Hao, Pan Li, Jinjun Xiong, and Deming Chen. ”Extensible and Efficient Proxy for Neural Architecture Search.” In IEEE/CVF International Conference on Computer Vision (ICCV), IEEE, 2023, 6176-6187.
  7. Bellodi, Elena, Davide Bertozzi, Alice Bizzarri, Michele Favalli, Michele Fraccaroli, and Riccardo Zese. ”Efficient Resource-Aware Neural Architecture Search with a Neuro-Symbolic Approach.” In IEEE 16th International Symposium on Embedded Multicore/Many-core Systems-on-Chip (MCSoC), IEEE, 2023, 171-178.
  8. Cassimon, Amber, Siegfried Mercelis, and Kevin Mets. ”Scalable reinforcement Learning-Based Neural Architecture Search.” Neural Computing and Applications, 2025, vol. 37, no. 1: 231-261.
  9. Zhao, Yiyang, Linnan Wang, and Tian Guo. ”Multi-objective Neural Architecture Search by Learning Search Space Partitions.” Journal of Machine Learning Research, 2024, vol. 25, no. 177: 1-41.
  10. Zhang, Yu-Ming, Jun-Wei Hsieh, Xin Li, Ming-Ching Chang, Chun-Chieh Lee, and Kuo-Chin Fan. ”MOTE-NAS: Multi-objective Training-Based Estimate for Efficient Neural Architecture Search.” Advances in Neural Information Processing Systems, 2024, vol. 37: 100845-100869.
  11. LeCun, Y., Cortes, C. and Burges, C.J.C. (1998) The MNIST Database of Handwritten Digits. New York, USA. http://yann.lecun.com/exdb/mnist/