University of Tokyo

05/06/2025 | Press release | Distributed by Public on 05/05/2025 09:08

Privacy-aware building automation

Split learning. The work of learning about the environment to better understand user intentions in order to control it is distributed between devices. ©2025 Ochiai et al. CC-BY-ND

Researchers at the University of Tokyo developed a framework to enable decentralized artificial intelligence-based building automation with a focus on privacy. The system enables AI-powered devices like cameras and interfaces to cooperate directly, using a new form of device-to-device communication. In doing so, it eliminates the need for central servers and thus the need for centralized data retention, often seen as a potential security weak point and risk to private data.

We live in an increasingly automated world. Cars, homes, factories and offices are gaining a range of automated functions to steer them, heat them, light them, or control them in some way. There are a number of approaches to automation systems, but at present most require a lot of programmed behaviors, which can be labor-intensive and inflexible, or when AI is involved, requires a high degree of centralization. But this brings with it some risk.

"A typical home or office automation system for lights or temperature control may involve cameras to monitor occupants and alter conditions on their behalf," said Associate Professor Hideya Ochiai from the Department of Information and Communication Engineering. "Under a conventional approach, such data, which most consider quite personal, especially if it's from your own home, will be aggregated on a central system. A breach of this system could risk leakage of that personal data. So my team and I devised an improved approach that is not only decentralized but also does away with the need to store personal data longer than is needed for the immediate automation processes to take place."

Their approach, called Distributed Logic-Free Building Automation (D-LFBA), describes how devices such as cameras and other sensors, and controllers for lights or temperature control, can be made to communicate directly, which avoids relying on centralization, and can be given a small amount of internal storage, mitigating the need to capture and keep more data than is necessary.

"We effectively spread the load of a neural network, the computer program responsible for learning and controlling things, across the devices in the environment," said Ochiai. "Among the advantages already mentioned, it should provide a cross-vendor layer of compatibility, meaning the automation environment need not be composed of systems from one manufacturer."

What makes D-LFBA especially unique is its ability to learn without being programmed. Using synchronized timestamps, the system matches images with corresponding control states over time. As users interact with their environment, by flipping switches or moving between rooms, the system learns those preferences. Over time, it adjusts automatically.

"Even without humans writing logic, the AI can generate fine-grained control," said Ochiai. "We saw that during trials last year, users were amazed at how well the system adapted to their habits."

Papers

Ryosuke Hara, Hiroshi Esaki, Hideya Ochiai, "Privacy-Aware Logic Free Building Automation Using Split Learning," IEEE Conference on Artificial Intelligence 2025: May 5, 2025

Related links

Inquiries about the content of this page: Public Relations Office

Send inquiry

University of Tokyo published this content on May 06, 2025, and is solely responsible for the information contained herein. Distributed via Public Technologies (PUBT), unedited and unaltered, on May 05, 2025 at 15:08 UTC. If you believe the information included in the content is inaccurate or outdated and requires editing or removal, please contact us at support@pubt.io