TinyML vs LLMs: A Survey of Extreme Scales in Machine Learning
Lists
Only you see this
Abstract
ML now works with an unmatched range of computational scales, from microcontrollers with models that are only a few kilobytes big to cloud infrastructures with Large Language Models (LLMs) that have trillions of parameters. These extremes show how AI has made progress in different ways: Tiny Machine Learning (TinyML) is great for green apps that don’t have a lot of resources because it focuses on real-time, ultra-low-power inference on edge devices. LLMs, on the other hand, are the best at understanding, reasoning, and generating language, but they often use a lot of energy and hurt the environment. This research provides a comprehensive analysis of TinyML and LLMs, highlighting their importance in the context of green and sustainable technology. We offer a systematic taxonomy that evaluates both paradigms based on critical factors such as model size, latency, energy consumption, memory footprint, training methodologies, privacy considerations, and deployment architecture. There are talks about how to use TinyML in smart agriculture, IoT, and wearable health monitoring, and how to use LLMs in cloud-based text generation, code synthesis, and decision support. We carefully consider the pros and cons of centralization and decentralization, scalability and sustainability, and performance and efficiency. Finally, we talk about what might happen in the future, like quantized TinyLLMs, federated learning for privacy-aware sustainability, and hybrid edge-cloud systems where TinyML acts as a green intelligence filter for LLM backends. The purpose of this poll is to get people to share their ideas on how to make AI systems that are both helpful and good for the environment.
BibTeX
@misc{lamaakal2025tinyml,
author = {Ismail Lamaakal and Chaymae Yahyati and Khalid El Makkaoui and Yassine Maleh and Ibrahim Ouahbi},
title = {{TinyML vs LLMs: A Survey of Extreme Scales in Machine Learning}},
year = {2025},
doi = {10.1109/icesa66763.2025.11281142},
}Loading related content...