Enhancing its tools to accelerate embedded Artificial-Intelligence (AI) and Machine-Learning (ML) development projects, STMicroelectronics has released upgrades to both NanoEdge AI Studio and STM32Cube.AI. These tools allow you to move AI and ML to an application’s edge. AI/ML offers significant advantages at the edge. These include privacy by design and deterministic and immediate response. They also offer greater reliability and lower power consumption.
NanoEdge AI Studio automates ML for applications that don’t require the creation of neural networks. It works with STM32 microcontrollers, MCUs, and MEMS sensors which include ST’s embedded intelligent sensor processing un (ISPU). STM32Cube.AI, an AI model optimizer for STM32, is available to developers who need to use neural networks. These two new releases provide features that allow you to quickly design and implement high-performance AI/ML systems with minimal investment. NanoEdge AI Studio version3.2 now includes an automatic datalogger generator to increase development productivity. It uses the ST development board as inputs and can also be configured with developer-defined parameters such as data rate and range, sample size and number of axes. These inputs allow NanoEdge AI Studio to generate the binary for the development boards without the need to code. The new features in NanoEdge AI Studio that allow data manipulation allows the user to optimize and clean up the data captured in NanoEdge AI Studio. This is important because it directly affects machine learning performance.
The validation stage, which allows users to evaluate their algorithms, shows inference time, memory use, and common performance metrics like accuracy and F1-Score. The validation stage also provides more details about the pre-processing used in the library and the ML model. NanoEdge AI Studio’s latest enhancement adds new pre-processing techniques, ML models and regression algorithms to improve performance. The tool also supports the creation of smart libraries that predict future system state using multi-order regression models. STM32Cube.AI version 7.3 can be used to develop cutting-edge AI/ML strategies. It is fully integrated in the STM32 ecosystem and allows conversion of pre-trained neural networks to optimized C code for industry’s most well-known 32-bit Arm(r] Cortex(r),-core MCUs. STM32Cube.AI has more flexibility for optimizing neural networks (NN). This tool adapts existing neural networks to meet performance requirements, fit within memory limits, or can do a balanced optimization that combines both. This update includes support for TensorFlow 2.10 models as well as new kernel performance improvements.
