A new approach using natural language models is expanding the use of artificial intelligence in edge computing. Using an advanced artificial intelligence (AI) technology, large data sets, such as ocean temperature, can be reconstructed with only a very small number of sensors deployed in the field. This approach leverages energy-efficient "edge" computing and has a wide range of potential uses in various fields including industry, scientific research and healthcare.
Javier Santos, a researcher at Los Alamos National Laboratory, said: "We have developed a neural network that allows us to represent a large system in a very compact way. This compactness means that it requires fewer computing resources than state-of-the-art convolutional neural network architectures, making it ideal for field deployment on drones, sensor arrays and other edge computing applications, bringing computing closer to its end use."
Novel artificial intelligence approach improves computational efficiency
Santos is the first author of a paper published in Nature Machine Intelligence by a team of Los Alamos researchers who call this new artificial intelligence technology Senseiver. This work is based on PerceiverIO, an artificial intelligence model developed by Google, and applies the technology of natural language models such as ChatGPT to the problem of reconstructing information about vast areas (such as oceans) from relatively small amounts of measurement data.
The research team realized that the model had broad application prospects due to its high efficiency. "Using fewer parameters and less memory requires fewer computer central processing unit cycles, so it runs faster on smaller computers," said paper co-author Dan O'Malley, a Los Alamos researcher.
In the first published validation of the model, Santos and his colleagues at Los Alamos demonstrated its effectiveness on real-world sparse data sets (meaning that information obtained from sensors covers only a very small portion of the area of interest) and on complex data sets of three-dimensional fluids.
In a demonstration of Senseiver's utility in the real world, the team applied the model to the National Oceanic and Atmospheric Administration's sea surface temperature data set. The model is able to integrate decades of vast amounts of measurement data obtained from satellites and shipboard sensors. From these sparse point measurements, the model predicts temperatures across the ocean, providing useful information for global climate models.
Bringing artificial intelligence to drones and sensor networks
Senseiver is well suited to the variety of projects and research areas of interest to Los Alamos.
"Los Alamos has extensive remote sensing capabilities, but using AI is not easy because the models are too large to fit on devices in the field, which leads to the need for edge computing," said Hari Viswanathan, a Los Alamos National Laboratory researcher, environmental scientist and co-author of the Senseiver paper. "Our work brings the benefits of AI to drones, field-based sensor networks, and other applications that are currently beyond the reach of cutting-edge AI technology."
AI models are particularly useful in the laboratory's work identifying and characterizing unmaintained wells. The laboratory leads the Department of Energy-funded Consortium to Assess Technology for Lost Oil and Gas Wells (CATALOG), a federal program tasked with identifying the location and characteristics of undocumented and unmaintained wells and measuring their methane emissions. Viswanathan is CATALOG's chief scientist.
This approach provides enhanced capabilities for large-scale practical applications such as autonomous vehicles, remote modeling of oil and gas assets, patient medical monitoring, cloud gaming, content delivery, and contaminant tracking.