Japanese electronics company TDK has announced advancements in sensor technology to optimise and accelerate the deployment of smart IoT.

Its SensorGPT technology uses generative AI, signal processing, statistical methods and simulations to create and manage sensor data at scale. This should empower both the smart IoT market and the emerging ambient IoT market segment to overcome key scalability problems. It streamlines model development and deployment, cutting time and cost, and enhancing the performance and efficiency of edge AI models and applications.

Data are the bedrock of intelligence in smart edge systems, yet data collection consumes more time than building the intelligence it is meant to power. Nearly 80% of AI development time is spent on data collection and curation. As the demand for edge AI continues to accelerate, projected to become the standard this year, data availability has become the primary barrier to scalability. SensorGPT directly addresses this by reducing reliance on real-world data through intelligent sensor data synthesis, cutting data collection efforts from 80% to nearly 10%, and enabling faster, more scalable edge AI development.

By using techniques to expand and enhance existing datasets, edge AI model building that takes months can be reduced to weeks, according to Jim Tran from TDK in the USA. “By tapping into generative AI modelling, simulation and more, engineers can use AI to generate additional, high-quality data that reflect real-world conditions, turning data into a scalable resource,” he said.

With SensorGPT data synthesis technology advancements, users can train generative models over limited real-world data to learn underlying patterns and generate high-quality synthetic data that faithfully mimic real-world data.

Users can leverage physics-based and mathematical models to simulate and generate synthetic sensor data. They can employ mathematical and computational techniques to simulate data reflecting the dynamics and characteristics of real sensor outputs. Data augmentation techniques automatically transform existing sensor data into rich, diverse datasets spanning a wide range of conditions and scenarios. Users can also streamline the labelling of training data, increasing its usefulness and quality for model training.

SensorGPT generates 90% similarity between synthetic and real-world sensor data, enabling the use of the synthetically generated data for faster edge AI deployment. Once deployed, it drives a virtuous cycle of feedback-driven improvement in which real-world data progressively refines and strengthens synthetic models over time, which in turn leads to more efficiently deployed models.

Users can improve scalability by generating large and diverse datasets that quickly help create AI for edge applications. It provides quick access to data for prototyping, testing and deploying initial models. Tools tailor data to specific sensors, smart IoT applications and real-world scenarios and conditions they operate in.

SensorGPT from TDK (www.tdk.com) can accelerate prototyping and proof of concepts, enabling orders-of-magnitude dataset size expansion, depending on the application and use case, reducing edge AI model building time from more than five months down to a few weeks.

Main applications include IoT, wearables, mobile, ambient IoT, industrial IoT and physical AI.