A large number of AI solution service providers participated in the 2023 International Artificial Intelligence Competition. Amid the exhibition of various image analysis and recognition/detection technologies used in computer vision, the AI development team was also blowing a wind of no-code development.

We met with CEO Du-Won Jeong of AltoAir, a no-code edge AI development platform service company that supports embedded AI development end-to-end.

■ Edge AI development faces problems of increased development period costs

The edge AI development process goes through the stages of △field data collection △data labeling △AI model learning △AI model lightweighting △AI model deployment, and it takes 3 to 4 months to complete one project. If issues such as △low accuracy, △lack of memory, △slow processing speed, or △labeling problems occur at this stage of the project, it is necessary to go back to the previous stage and perform a redesign, which leads to an increase in development time and cost.

AltoAir CEO Jeong Du-won cited three reasons why AI models return to the design stage: △ difficulty in predicting the memory usage of design models, △ hardware constraints used in model design, and △ lack of know-how to confirm optimization of AI models in terms of accuracy and processing speed.

In particular, edge AI generally has constraints in processing power, memory, storage capacity, etc., so the AI model process necessarily involves the development of a 'fit' model optimized for these constraints, and the optimization process involves numerous trials and errors, repeated redesigns, and modifications.

Representative Jeong emphasized that since computational volume, processing speed, and memory usage are in a trade-off relationship, it is most important to appropriately distribute the load between SW and HW and find a balance point. Accordingly, when individual developers design AI models on their own, the time, energy, and resources invested in a single model training are bound to be excessive.

■ AltoAir Launches No-Code Edge AI Development Platform

AltoAir Inc. ambitiously introduced its no-code edge AI development platform, ‘Tiny Boom’, at this year’s International Artificial Intelligence Competition.

Most companies provide AI solution services for each technology and often provide customized technical responses to customers. On the other hand, through the AI development platform, the entire process from data collection to AI model learning and lightweighting, and AI model distribution can be executed in one stop using an intuitive UI/UX.

CEO Jeong Du-won, who developed TinyBoom, emphasized, “This is a product that allows even companies with no AI capabilities to develop edge AI models with simple UI manipulation,” and “It can reduce development time by more than 1/10 of the existing time, and it can also realize the effect of improving the efficiency of AI expert engineer resources.”

The types of AI models that can be developed through Tiny Boom are divided into image data analysis models such as image classification and object recognition, and time series data analysis models such as signal classification and anomaly detection. CEO Jeong added that there are cases where it has been supplied to quality inspection processes at steel manufacturing companies and medical image interpretation.

CEO Jeong emphasized, “Tinyboom is a solution specialized in the development of edge AI models that are relatively inexpensive enough to be operated on microcontrollers (MCUs) and can be operated on low-power processors,” and said, “In the second half of 2023, we plan to provide a ‘HW-NAS’ function to minimize the edge AI development cycle by applying technology that finds the optimal balance between HW and SW.”