NEC Begins Building an AI Foundation Model for Underwater Acoustics

Japanese technology company NEC Corporation has begun research on a foundation model -- a general-purpose artificial intelligence (AI) system trained on large volumes of data -- for underwater acoustics, after winning a contract from Japan's Acquisition, Technology and Logistics Agency, the government body responsible for the country's defense procurement and technology development. Announced on August 18, the project falls under the agency's Innovative Breakthrough Research program and is designed for dual-use applications spanning both defense and civilian needs, including marine life and resource exploration and environmental monitoring.

NEC aims to build the model by fiscal 2027, training it on large volumes of underwater acoustic data collected through public-private collaboration to create a general-purpose system for underwater sound, similar to how large language models underpin generative AI for text. The company said the model is intended to improve the speed and accuracy of identifying underwater events that have been difficult to detect with existing methods, and could support the creation of a digital twin of the ocean -- a detailed virtual model of ocean conditions.

Underwater observation relies heavily on sound because light and radio waves travel only short distances through water. Acoustic data gathered through sonar and hydrophones typically require analysis by specialists using signal-processing techniques, but NEC said the sheer volume of data makes analysis time-consuming and limits achievable accuracy. A broad-based foundation model for underwater acoustics has not yet been widely realized, the company said, with existing AI systems largely limited to models built for specific tasks and requiring large amounts of labeled training data that are difficult to obtain underwater.

NEC said it will work with Japan's Defense Innovation Science and Technology Institute to develop and verify the model, drawing on more than 90 years of sonar technology development and the large-scale training expertise behind its in-house AI system, cotomi; potential applications the company cited include marine life and resource exploration, environmental monitoring and high-precision earthquake forecasting.