aiqbench analysis
CoLoRA extends the popular LoRA technique to convolutional neural networks by decomposing kernel updates, achieving over 80% reduction in trainable parameters while maintaining inference efficiency. The study validates this approach on medical imaging tasks like OCT classification and standard benchmarks, highlighting that parameter efficiency does not automatically guarantee proportional memory savings during training. This method offers a practical pathway for deploying large vision models with constrained compute resources without altering the original architecture.
COLORA: Efficient Fine-Tuning for Convolutional Models with a Study Case on Optical Coherence Tomography Image Classification
arXiv:2505.18315v4 Announce Type: replace-cross Abstract: We introduce CoLoRA (Convolutional Low-Rank Adaptation), a parameter-efficient fine-tuning method for convolutional neural networks (CNNs). CoLoRA extends LoRA to convolutional layers by decomposing kernel updates into lightweight depthwise and pointwise components. This design reduces the number of trainable convolutional-update parameters by over 80\% compared with full convolutional fine-tuning, while allowing the learned updates to be merged into the pretrained convolutional kernels, thereby preserving the original model size and inference complexity. Experiments on MedMNIST datasets, particularly OCTMNISTv2, demonstrate that CoLoRA applied to VGG16 and ResNet50 achieves competitive classification performance while substantially reducing the number of trainable parameters. Comparisons with transfer learning, adapters, BitFit, and convolutional LoRA variants further characterize the trade-offs among predictive performance, trainable parameters, and training cost. Additional experiments on CIFAR-100 and Cats vs. Dogs provide preliminary evidence that the proposed adaptation strategy also transfers to non-medical image-classification tasks. Peak GPU-memory measurements further show that parameter efficiency does not translate directly into proportional training-memory savings, with memory consumption depending strongly on the placement of the adapted convolutional layers. Overall, CoLoRA provides a parameter-efficient and deployment-efficient alternative to full fine-tuning for convolutional models.
Abstract:We introduce CoLoRA (Convolutional Low-Rank Adaptation), a parameter-efficient fine-tuning method for convolutional neural networks (CNNs). CoLoRA extends LoRA to convolutional layers by decomposing kernel updates into lightweight depthwise and pointwise components. This design reduces the number of trainable convolutional-update parameters by over 80\% compared with full convolutional fine-tuning, while allowing the learned updates to be merged into the pretrained convolutional kernels, thereby preserving the original model size and inference complexity. Experiments on MedMNIST datasets, particularly OCTMNISTv2, demonstrate that CoLoRA applied to VGG16 and ResNet50 achieves competitive classification performance while substantially reducing the number of trainable parameters. Comparisons with transfer learning, adapters, BitFit, and convolutional LoRA variants further characterize the trade-offs among predictive performance, trainable parameters, and training cost. Additional experiments on CIFAR-100 and Cats vs. Dogs provide preliminary evidence that the proposed adaptation strategy also transfers to non-medical image-classification tasks. Peak GPU-memory measurements further show that parameter efficiency does not translate directly into proportional training-memory savings, with memory consumption depending strongly on the placement of the adapted convolutional layers. Overall, CoLoRA provides a parameter-efficient and deployment-efficient alternative to full fine-tuning for convolutional models.
Computer Vision and Pattern Recognition (cs.CV); Artificial Intelligence (cs.AI)
I.1.2; I.4.0; I.4.10; I.4.0
arXiv:2505.18315 [cs.CV]
(or arXiv:2505.18315v4 [cs.CV] for this version)
https://doi.org/10.48550/arXiv.2505.18315 Focus to learn more arXiv-issued DOI via DataCite
arXivLabs is a framework that allows collaborators to develop and share new arXiv features directly on our website.
Both individuals and organizations that work with arXivLabs have embraced and accepted our values of openness, community, excellence, and user data privacy. arXiv is committed to these values and only works with partners that adhere to them.
Have an idea for a project that will add value for arXiv's community? Learn more about arXivLabs.
Full text under the publisher's open licence · 285 words
Related coverage
Transferable Low-Rank Convolutional Bases for Onboarding Unseen Medical Imaging Modalities
arXiv · Jul 21, 2026
Quantum-Enhanced Learning Framework for Intelligent and AI-Native 6G Wireless Networks
arXiv · Sep 9, 2026
Links #5: 2026/07
LessWrong · Jul 31, 2026
Too much of a good thing -- when knowledge distillation promotes overfitting, and how to avoid it
arXiv · Aug 27, 2026