In recent years, deep learning has become popular in research due to its applicability in many industries. For instance, deep learning can be applied in healthcare, security surveillance, self-drive car, human activity recognition, recommended systems, image quality enhancement, transportation, prediction, forecasting, etc. Before the introduction of deep learning, prediction and decision-making can be achieved using statistical methods and machine learning. Deep learning algorithms have been solving successfully complex real-time problems. This was previously not possible with machine learning and computer vision methods.

Deep learning has also gained popularity because it automatically extracts the important features from training data and these features help to make an appropriate decision. However, there are challenges in deep learning such as problems of high computation power and resources. Moreover, deep learning models are computationally extensive and require high storage space. Therefore, a deep learning model is not well suited for edge devices. Users are not able to get high computation resources in a real-time domain from a remote location or in the case of mobility. Hence, these deep learning models require significant improvement. For instance, there is a need to make deep learning models that include lightweight and better inference time so that the models can be compatible with resource-constrained devices. Recent research has shown significant improvement in compression techniques by applying pruning, lossy weight encoding, parameter sharing, multilayer pruning, low-rank factorization, etc. To compress deep learning models, two approaches exist: compression during training and compression of the trained model. Moreover, various techniques are available for model optimization and compression for resource-constrained devices. For instance, genetic algorithms, swarm optimization, swarm intelligence, nature-inspired optimization, game-theoretic approach, chemical reaction optimization, and differential evolution.

The aim of the Special Issue is to bring together original research and review articles discussing the compression of deep learning models for resource-constrained devices. We welcome submissions from researchers who have been working on the development and deployment of deep learning models in edge devices (e.g., Raspberry Pi, Google edge tensor processing unit (Google TPU), NVIDIA Jetson Nano Developer Kit, Android devices, etc). This Special Issue invites original research discussing innovative architectures and training methods for effective and efficient compression.

Potential topics include but are not limited to the following:

• Model optimization and compression for medical applications
• Model optimization and compression for Internet of Things (IoT) and edge applications
• Model optimization and compression for deep learning algorithms in security analysis applications
• New architectures for model compression include pruning, quantization, knowledge distillation, neural architecture search (NAS), etc.
• Generalize lightweight architectures for deep learning problems
• Compression approaches for deep reinforcement learning
• Efficient use of computation resources for executing deep learning models
• Architectures and models that work with less training data on remote applications
• Compressed deep learning model for explainable artificial intelligence
• Compressed and accelerated versions of famous pr-trained architectures (e.g., AlexNet, OxfordNet (VGG16), residual neural network (ResNet), etc.)
• Compression and acceleration of object detection model such as "You Only Look Once" (YOLO) model, and single-shot detector (SSD)
• Accelerate the UNet and VNet architectures
• Methodology and framework for developing storage constraints in deep learning models