Shrink-Wrapping Deep Learning: A Curated Guide to TinyML

What it does

This repository is a curated markdown bibliography of papers, projects, articles, and talks focused on TinyML—the practice of running machine learning models on microcontrollers and other resource-starved devices. It organizes research chronologically from 2016 through 2025 and tags foundational work like SqueezeNet, CMSIS-NN, and TensorFlow Lite Micro alongside newer entries in neural architecture search and wake-word detection. Think of it as a field guide for developers who need to know which ideas actually work when you have no GPU and barely any RAM.

The interesting bit

The list’s real value is in its specificity: it doesn’t just collect “edge AI” fluff, but surfaces papers on exacting constraints like integer-only quantization, Kronecker-product compression, and RNN inference in 2 KB of RAM. That granularity turns a generic literature search into a targeted scavenger hunt for proven techniques.

Key highlights

• Chronological coverage from 2016 to 2025, showing how the field evolved from basic pruning to hardware-aware NAS.
• Explicit links to PDFs and official source code for many entries, so you can read the paper and then inspect the implementation.
• Sections for benchmarking, courses, books, and tools, making it closer to a syllabus than a simple link dump.
• Includes niche but practical topics like wake-word spotting, neuromorphic vision sensors, and federated learning on tiny devices.

Caveats

• The README is a flat list; there is no taxonomy by technique (e.g., quantization vs. pruning) or by hardware target, so you have to know what year a paper came out to browse chronologically.
• Most entries are just titles and PDF links with no annotation, so you’ll still need to skim abstracts to judge relevance.

Verdict

Grab this if you are building on-device ML and need a cheat sheet to the seminal papers; skip it if you are looking for a framework, benchmark suite, or tutorial code.