Advancing Generative AI techniques
TDK Advances Edge AI Development With SensorGPT
TDK has introduced SensorGPT, a technology designed to accelerate the development and deployment of AI-based IoT applications at the edge.
The technology uses generative AI, signal processing, statistical methods and simulations to create and manage synthetic sensor data at scale.
According to TDK, SensorGPT is intended to address scalability challenges in smart IoT and Ambient IoT applications by reducing the dependency on real-world sensor data during AI model development.
Reduced reliance on real-world data
The company states that up to 80 percent of AI solution development time is currently spent on collecting and curating data. SensorGPT is designed to reduce the need for real-world data from around 80 percent to approximately 10 percent by generating synthetic datasets that reflect operational conditions.
“By using advanced techniques to expand and enhance existing datasets, edge AI model building time that takes months can be reduced to weeks,” said Jim Tran, Corporate Officer and General Manager, Americas HQ and Deputy General Manager, Technology & Intellectual Property HQ, TDK USA Corporation.
According to TDK, the generated synthetic data achieves up to 90 percent similarity with real-world sensor data.
Combination of AI and simulation methods
SensorGPT uses several technologies for data synthesis. These include generative AI models trained on limited real-world datasets, physics-based simulation models and signal-processing methods that recreate sensor behavior mathematically.
SensorGPT also supports automated data augmentation and assisted annotation to simplify the labeling of training data.
TDK says the technology can improve the scalability of edge AI applications by generating larger and more diverse datasets for training and testing. The company also highlights faster prototyping cycles and reduced development costs.
Focus on IoT and industrial applications
The technology targets applications in IoT, wearables, mobile devices, Ambient IoT and Industrial IoT environments. TDK also cites physical AI applications as a potential field of use.
According to the company, SensorGPT can shorten edge AI model development cycles from more than five months to a few weeks, depending on the application scenario.
