Ahmed Radwan
Current Organizations / Institutes
The organizations/institutes from the scholar's latest article
Research Subjects
Ahmed Radwan has a total of 7 co-authors. Here are the top co-authors by number of shared publications:
- Mahmoud Soliman
The University of British Columbia
3 shared publications - Mohamed Sami Shehata
The University of British Columbia
3 shared publications - Omar Abdelaziz
The University of British Columbia
3 shared publications - Ahmed Elgazwy
The University of British Columbia
2 shared publications - Ahmad Abdel-Qader
School of Engineering, The University of British Columbia, Kelowna, BC, Canada
1 shared publication Mohamed S. Shehata
1 shared publication
Latest Publications
- 3 November 2026
- Published by Elsevier in Computer Vision and Image Understanding
- Vol. 274
SODA: Stabilized Optimal Transport based Domain Alignment for Domain Generalization via a Dynamic Feature Reservoir
- 3 November 2026
- Published by Elsevier in Computer Vision and Image Understanding
- Vol. 274
SODA: Stabilized Optimal Transport based Domain Alignment for Domain Generalization via a Dynamic Feature Reservoir
- 4 August 2026
- Published by Springer Nature
- p. 182-197
MetaMoA: Top-Down Dynamic Guidance for Parameter-Efficient Domain Generalization
- 4 August 2026
- Published by Springer Nature
- p. 182-197
MetaMoA: Top-Down Dynamic Guidance for Parameter-Efficient Domain Generalization
- 4 August 2026
- Published by Springer Nature
- p. 349-363
FedDG-MoE-NF: Prototypical Normalizing Flow Networks for Federated Domain Generalization
- 4 August 2026
- Published by Springer Nature
- p. 349-363
FedDG-MoE-NF: Prototypical Normalizing Flow Networks for Federated Domain Generalization
- Open Access
- 8 July 2026
- Published by Elsevier in SSRN Electronic Journal
SVDTrack: Parameter-Efficient Tracking via Residual-Subspace Adaptation
Deploying transformer-based single object trackers on resource-constrained devices remains challenging because adapting a pre-trained backbone conventionally requires fine-tuning on the order of 90M parameters. Parameter-Efficient Fine-Tuning (PEFT) mitigates this cost, yet existing PEFT trackers often trade tracking accuracy for efficiency. We introduce SVDTrack, a PEFT method that uses the Singular Value Decomposition (SVD) of the pre-trained attention and MLP weight matrices to decide where adaptation should occur. SVDTrack combines two targeted modifications: (1) a learnable reweighting of the singular values, S = w ⊙ Σ + b, and (2) low-rank adaptation restricted to the singular vectors associated with the least significant singular values. We show that any update confined to this residual singular subspace provably preserves the optimal rank-k approximation of the pre-trained weights—the model’s core knowledge—and we introduce a constrained Residual LoRA variant that satisfies this condition by construction, which is essential when adapting weights already fine-tuned on tracking data. After training, all learned modifications are reparameterized back into standard weight matrices, so the deployed model incurs zero additional inference latency or memory. On four benchmarks (LaSOT, LaSOText, TrackingNet, GOT-10k), SVDTrack sets a new state of the art among PEFT trackers and matches or surpasses fully fine-tuned models on several metrics, while training only 9.57M parameters—an 89.6% reduction relative to full fine-tuning. Controlled ablations show that adapting the residual subspace outperforms adapting the most significant or randomly selected components under a matched parameter budget, supporting principled, SVD-guided adaptation as a practical route to efficient, high-performance tracking on edge devices.
- 8 July 2026
- Published by Elsevier in SSRN Electronic Journal
Open AccessSVDTrack: Parameter-Efficient Tracking via Residual-Subspace Adaptation
Deploying transformer-based single object trackers on resource-constrained devices remains challenging because adapting a pre-trained backbone conventionally requires fine-tuning on the order of 90M parameters. Parameter-Efficient Fine-Tuning (PEFT) mitigates this cost, yet existing PEFT trackers often trade tracking accuracy for efficiency. We introduce SVDTrack, a PEFT method that uses the Singular Value Decomposition (SVD) of the pre-trained attention and MLP weight matrices to decide where adaptation should occur. SVDTrack combines two targeted modifications: (1) a learnable reweighting of the singular values, S = w ⊙ Σ + b, and (2) low-rank adaptation restricted to the singular vectors associated with the least significant singular values. We show that any update confined to this residual singular subspace provably preserves the optimal rank-k approximation of the pre-trained weights—the model’s core knowledge—and we introduce a constrained Residual LoRA variant that satisfies this condition by construction, which is essential when adapting weights already fine-tuned on tracking data. After training, all learned modifications are reparameterized back into standard weight matrices, so the deployed model incurs zero additional inference latency or memory. On four benchmarks (LaSOT, LaSOText, TrackingNet, GOT-10k), SVDTrack sets a new state of the art among PEFT trackers and matches or surpasses fully fine-tuned models on several metrics, while training only 9.57M parameters—an 89.6% reduction relative to full fine-tuning. Controlled ablations show that adapting the residual subspace outperforms adapting the most significant or randomly selected components under a matched parameter budget, supporting principled, SVD-guided adaptation as a practical route to efficient, high-performance tracking on edge devices.