NTT has developed the world’s first deep learning AI algorithm “test-time adaptation technology” that autonomously adapts a numerical prediction model (regression model) to environmental changes during learning and operation.

This result makes it possible for a trained regression model to autonomously adapt using only unsupervised data obtained from the operating environment when it is placed in the operating environment.

Despite its high practicality, a test-time adaptation technology for regression models has been established, which has not been researched until now. Since it does not depend on the model output format, by incorporating it into data analysis AI that is used in various business fields including manufacturing, medical care, and finance, it is possible to prevent accuracy degradation due to environmental changes, and it is expected to lead to significant cost reductions in MLOps. In addition, this knowledge can be applied to tasks other than regression tasks in response to environmental changes such as weather changes and sensor deterioration in images and numerical data in multimodal platform models.

※Translating Japanese articles into English with AI