Model Compression & Optimization
Model compression has emerged as an important area of research for deploying deep learning models on IoT devices. However, model compression is not a sufficient solution to fit the models within the memory of a single device; as a result we need to distribute them across multiple devices. This leads to a distributed inference paradigm in which communication costs represent another major bottleneck. To this end, we focus on knowledge distillation and ‘teacher’ – ‘student’ type of architectures for distributed model compression, as well as data independent model compression.
Selected Publications
Munir, Mustafa; Zhang, Alex; Marculescu, Radu
VCMamba: Bridging Convolutions with Multi-Directional Mamba for Efficient Visual Representation Conference
Proceedings of the International Conference on Computer Vision (ICCV 2025) Workshops, 2025.
@conference{VCMamba_ICCV_2025,
title = {VCMamba: Bridging Convolutions with Multi-Directional Mamba for Efficient Visual Representation},
author = {Mustafa Munir and Alex Zhang and Radu Marculescu},
url = {https://arxiv.org/abs/2509.04669},
year = {2025},
date = {2025-10-19},
urldate = {2025-10-19},
publisher = {Proceedings of the International Conference on Computer Vision (ICCV 2025) Workshops},
abstract = {Recent advances in Vision Transformers (ViTs) and State Space Models (SSMs) have challenged the dominance of Convolutional Neural Networks (CNNs) in computer vision. ViTs excel at capturing global context, and SSMs like Mamba offer linear complexity for long sequences, yet they do not capture fine-grained local features as effectively as CNNs. Conversely, CNNs possess strong inductive biases for local features but lack the global reasoning capabilities of transformers and Mamba. To bridge this gap, we introduce VCMamba, a novel vision backbone that integrates the strengths of CNNs and multi-directional Mamba SSMs. VCMamba employs a convolutional stem and a hierarchical structure with convolutional blocks in its early stages to extract rich local features. These convolutional blocks are then processed by later stages incorporating multi-directional Mamba blocks designed to efficiently model long-range dependencies and global context. This hybrid design allows for superior feature representation while maintaining linear complexity with respect to image resolution. We demonstrate VCMamba’s effectiveness through extensive experiments on ImageNet-1K classification and ADE20K semantic segmentation. Our VCMamba-B achieves 82.6% top-1 accuracy on ImageNet-1K, surpassing PlainMamba-L3 by 0.3% with 37% fewer parameters, and outperforming Vision GNN-B by 0.3% with 64% fewer parameters. Furthermore, VCMamba-B obtains 47.1 mIoU on ADE20K, exceeding EfficientFormer-L7 by 2.0 mIoU while utilizing 62% fewer parameters. Code is available at https://github.com/Wertyuui345/VCMamba.},
keywords = {},
pubstate = {published},
tppubtype = {conference}
}
Munir, Mustafa; Li, Guihong; Rahman, Md Mostafijur; Zhang, Alex; Marculescu, Radu
From Data to Design: Leveraging Frequency Statistics for Efficient Neural Network Architectures Conference
2025 Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR) Workshops, 2025.
@conference{Data_to_Design_Frequency,
title = {From Data to Design: Leveraging Frequency Statistics for Efficient Neural Network Architectures},
author = {Mustafa Munir and Guihong Li and Md Mostafijur Rahman and Alex Zhang and Radu Marculescu},
url = {https://openaccess.thecvf.com/content/CVPR2025W/eLVM/html/Munir_From_Data_to_Design_Leveraging_Frequency_Statistics_for_Efficient_Neural_CVPRW_2025_paper.html},
year = {2025},
date = {2025-06-11},
urldate = {2025-06-11},
booktitle = {2025 Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR) Workshops},
keywords = {},
pubstate = {published},
tppubtype = {conference}
}
Munir, Mustafa; Rahman, Md Mostafijur; Marculescu, Radu
RapidNet: Multi-Level Dilated Convolution Based Mobile Backbone Conference
2025 Proceedings of the IEEE/CVF Winter Conference on Applications of Computer Vision (WACV 2025), 2025.
@conference{RapidNet_WACV_2025,
title = {RapidNet: Multi-Level Dilated Convolution Based Mobile Backbone},
author = {Mustafa Munir and Md Mostafijur Rahman and Radu Marculescu},
url = {https://ieeexplore.ieee.org/document/10943953},
year = {2025},
date = {2025-03-03},
urldate = {2025-03-03},
booktitle = {2025 Proceedings of the IEEE/CVF Winter Conference on Applications of Computer Vision (WACV 2025)},
journal = {2025 Proceedings of the IEEE/CVF Winter Conference on Applications of Computer Vision (WACV 2025)},
keywords = {},
pubstate = {published},
tppubtype = {conference}
}
Li, Guihong; Bhardwaj, Kartikeya; Yang, Yuedong; Marculescu, Radu
TIPS: Topologically Important Path Sampling for Anytime Neural Networks Conference
International Conference on Machine Learning (ICML), 2023.
@conference{tips_icml2023,
title = {TIPS: Topologically Important Path Sampling for Anytime Neural Networks},
author = {Li, Guihong and Bhardwaj, Kartikeya and Yang, Yuedong and Marculescu, Radu},
url = {https://arxiv.org/abs/2305.08021},
year = {2023},
date = {2023-07-15},
urldate = {2023-07-15},
booktitle = {International Conference on Machine Learning (ICML)},
keywords = {},
pubstate = {published},
tppubtype = {conference}
}
Goksoy, Alper-A; Li, Guihong; Mandal, Sumit K.; Ogras, Umit Y.; Marculescu, Radu
CANNON: Communication-Aware Sparse Neural Network Optimization Journal Article
In: IEEE Transactions on Emerging Topics in Computing, 2023.
@article{nokey,
title = {CANNON: Communication-Aware Sparse Neural Network Optimization},
author = {Alper-A Goksoy and Guihong Li and Sumit K. Mandal and Umit Y. Ogras and Radu Marculescu},
url = {https://ieeexplore.ieee.org/document/10171170},
year = {2023},
date = {2023-06-23},
urldate = {2023-06-23},
journal = {IEEE Transactions on Emerging Topics in Computing},
keywords = {},
pubstate = {published},
tppubtype = {article}
}
Hoang, Duc N. M; Liu, Shiwei; Marculescu, Radu; Wang, Zhangyang
Revisiting Pruning at Initialization through the Lens of Ramanujan Graph Conference
International Conference on Learning Representations (ICLR), 2023.
@conference{hoangiclr2023,
title = {Revisiting Pruning at Initialization through the Lens of Ramanujan Graph},
author = {Duc N.M Hoang and Shiwei Liu and Radu Marculescu and Zhangyang Wang},
url = {https://openreview.net/pdf?id=uVcDssQff_},
year = {2023},
date = {2023-01-25},
urldate = {2023-01-25},
booktitle = {International Conference on Learning Representations (ICLR)},
keywords = {},
pubstate = {published},
tppubtype = {conference}
}
