Constrained Linear Data-feature Mapping for Image Classification
This work addresses interpretability and efficiency for image classification researchers, but it is incremental as it builds on existing ResNet models.
The paper tackles the problem of interpretability and parameter efficiency in image classification by proposing a constrained linear data-feature mapping model, linking it to ResNet architectures, and showing that modified models achieve similar accuracy with fewer parameters.
In this paper, we propose a constrained linear data-feature mapping model as an interpretable mathematical model for image classification using convolutional neural network (CNN) such as the ResNet. From this viewpoint, we establish the detailed connections in a technical level between the traditional iterative schemes for constrained linear system and the architecture for the basic blocks of ResNet. Under these connections, we propose some natural modifications of ResNet type models which will have less parameters but still maintain almost the same accuracy as these corresponding original models. Some numerical experiments are shown to demonstrate the validity of this constrained learning data-feature mapping assumption.