This repository is a curated collection of research resources, papers, and libraries focused on embedded and mobile deep learning (EMDL). It serves as a comprehensive knowledge base for researchers and developers working on optimizing deep learning models for resource-constrained devices, aiming to make AI more accessible and efficient on edge hardware.

How It Works

The collection is organized thematically, covering key areas of EMDL research such as model compression (pruning, quantization, approximation), efficient architectures, system benchmarking, and hardware acceleration. It highlights seminal papers and practical libraries, providing a structured overview of the state-of-the-art techniques and tools for developing and deploying deep learning models on embedded and mobile platforms.

Quick Start & Requirements

This is a curated list of resources, not a runnable project. No installation or specific requirements are applicable.

Highlighted Details

• Extensive coverage of model compression techniques including pruning, quantization, and approximation.
• Detailed listings of libraries and frameworks for EMDL, such as TensorFlow Lite, PyTorch Mobile, Arm NN, and ONNX Runtime.
• Inclusion of system-level research on benchmarking, characterization, and hardware acceleration for embedded AI.
• Links to numerous academic papers and their corresponding repositories, offering deep dives into specific EMDL topics.

Maintenance & Community

The repository is maintained by csarron. It aggregates research from various institutions and companies, including MIT, Google, Arm, Huawei, and IBM, indicating a broad community interest.

Licensing & Compatibility

The repository itself is a list of links and does not have a specific license. The licenses of the linked papers and libraries vary, and users should consult the individual project licenses for compatibility and usage terms.

Limitations & Caveats

This is a research aggregation and does not provide a unified framework or runnable code. Users will need to explore and integrate individual libraries and techniques based on their specific project needs.