About DNA  Research Lab

IoT environments are diverse and dynamic — data distributions shift, labeled data is scarce, and building new models for every device or scenario is not scalable. AutoML addresses the scalability challenge by automating model design, while Meta-Learning addresses the data scarcity challenge by enabling rapid adaptation to new environments with only a handful of examples.

By utilizing millimeter-wave (mmWave) radar, a non-contact sensing technology, it is possible to assess human condition and detect surrounding environments. By emitting and analyzing high-frequency radio signals, minute body movements can be detected with remarkable precision. This enables the real-time measurement of various human physiological and behavioral signals. Key research areas are as follows:

• Vital Signs Monitoring: Remotely measuring heart rate and respiration rate without any body-worn sensors.

• Hand Gesture Recognition: Decoding intricate hand and finger movements to enable intuitive, touchless device control.

• Activity Monitoring: Classifying various daily activities and detecting significant events like falls for safety and care.

The gap between state-of-the-art model capability and real-world deployability is one of the central challenges in modern AI. IoT and edge devices — embedded controllers, microcontrollers, smart sensors — operate under strict constraints on memory, power, and latency. A model that performs well on a GPU cluster is meaningless if it cannot run on the device where decisions actually need to be made.

Optimization and light-weighting bridge this gap. They allow powerful models to be embedded directly into the physical world — enabling real-time visual inspection on a factory line, on-device language understanding in a smart home system, or generative inference without cloud dependency. As AI becomes infrastructure, the ability to make it efficient, portable, and deployable becomes as important as making it accurate.