Adaptive machine learning for communication-constrained edge devices — NSF AI grant $1.2M (CAML, NeTS)

The NSF awarded about $1,200,000 to research communication-constrained adaptive machine learning (CAML) for heterogeneous edge networks (NSF Award 2504762; program: Networking Technology and Systems [NeTS]; Nevada).

1The communication limits AIoT faces

Per the abstract, the project investigates principles and methodologies for AIoT (Artificial Intelligence of Things) devices to make decisions quickly and accurately — meeting safety, accuracy, and latency requirements — even with limited communication and computing power.

By developing machine-learning techniques that work well under communication-resource limitations, it aims to spur a new line of thinking for AIoT applications facing communication-constrained ML, such as smart health, connected cars, augmented reality, and smart city, benefiting society.

It also contributes to workforce development by integrating findings into undergraduate and graduate education and summer programs on data science, AI, and ML, broadening K-12 participation.

2Adapting to each device's bandwidth and compute

Technically, it develops Communication-constrained Adaptive Machine Learning (CAML) methods that adapt to varying communication bandwidth, computational power, and data size of edge devices, enabling heterogeneous devices to work in concert. It studies CAML for two popular edge-network settings: edge networks with a central coordinator and fully decentralized edge networks.

For the central-server setting, it studies distributed fine-tuning for on-device machine learning.

3Four in ten obligations land on a round ten thousand

Of the 120 NSF AI-related awards this site holds as of 2026-09-01, 49 carry an obligation that lands on a round ten thousand dollars.

Among federal procurement contracts, only eight of 222 landed on a round ten thousand. Among awards it is more than four in ten. Procurement is settled by adding up the cost lines a job requires, while an award is often placed against the frame or ceiling of a solicitation — and the difference shows in the shape of the numbers.