ETRI Launches Open Source No-Code Machine Learning Tools

The Electronics and Telecommunications Research Institute (ETRI) has unveiled a significant advancement in machine learning technology. On November 6, 2023, ETRI announced the release of its core technology for MLOps tools as open source on GitHub. This innovation is designed to facilitate the development of artificial intelligence (AI) applications by users with limited coding skills across various industries, including manufacturing and healthcare.

At a public seminar held at the Science and Technology Center in Gangnam-gu, Seoul, the research team showcased the TANGO framework. This technology automates the generation of neural networks without requiring extensive programming knowledge. It also streamlines the deployment process across diverse hardware environments, such as cloud systems and on-premise servers.

TANGO is particularly valuable in sectors where AI integration has proven challenging. For instance, while quality inspections in steel mills can identify defective steel, implementing AI solutions has been complex. Similarly, doctors often diagnose conditions like tuberculosis from X-ray images, yet developing AI models to assist in these diagnoses has been difficult. The TANGO framework addresses these challenges, making neural network processing accessible to domain experts who lack deep technical expertise.

Traditionally, domain experts were responsible for tasks like data labeling, while software developers focused on model development and application deployment. With the rising demand for AI solutions across industries, a shortage of qualified AI and software specialists has emerged. ETRI’s new tools aim to bridge this gap by offering an easily installable solution that can be quickly operationalized via a web interface.

In addition to TANGO, ETRI introduced LLMOps tools, which are aimed at supporting the development of generative AI technologies. The development of TANGO has resulted in a remarkable output, including 24 patents, three papers presented at the NeurIPS conference, and 13 publications in scientific journals. The initiative has also generated approximately KRW 10 billion in commercialization revenue.

One notable success story is Avenotics, a company specializing in autonomous maritime navigation solutions. It has been selected for the “Public Research Outcome Expansion and Commercialization Project,” backed by the Ministry of Science and ICT. Through technology transfer from ETRI, Avenotics has secured funding of 1.3 billion won from various investors, including Korea Science and Technology Holdings, Korea Credit Guarantee Fund, and Low Partners. The company is using Tango technology to enhance on-device AI capabilities, enabling the automatic generation of contextual information for navigators.

ETRI’s research team is committed to promoting broader adoption of these technologies. Collaborative efforts with institutions like Weda Co., Ltd. have led to the development of AI services for on-site employees in sectors such as steel and automotive manufacturing. Weda Co., Ltd.’s service focuses on vision-based inspection for complex shapes, like automotive bumpers.

Furthermore, Lablup Inc. has partnered with KT Cloud to create a deployment optimization service tailored for Rebellion’s domestic AI acceleration engine, ATOM-Max. This collaboration has also resulted in the commercialization of a GPU cloud rental service.

In the healthcare sector, Seoul National University Hospital is leveraging AI to analyze large-scale chest CT images and generate diagnostic reports. This technology is set to be tested in four hospitals, including Seoul National University Hospital and Boramae Hospital, to enhance cardiopulmonary disease prediction services.

The LLMOps tools, designed for generative AI, are progressing towards commercialization in collaboration with Acryl Inc. The source code for Acryl’s commercial product, Jonathan, is now publicly available on GitHub, with ongoing enhancements to core algorithms and the establishment of a standard operating environment for industry-specific applications.

Kim Tae-ho, Software Project Manager at the Institute of Information & Communications Technology Planning & Evaluation (IITP), praised TANGO as a leading open-source project in Korea, significantly bolstering the nation’s software industry in AI development tools. Meanwhile, Jo Chang Sik, Principal Researcher at ETRI, expressed intentions to expand TANGO’s capabilities, particularly in supporting generative AI through LLMOps tools.

ETRI plans to update the source code on GitHub every six months and host annual public seminars to share technological advancements and practical expertise. The first four TANGO seminars attracted 944 participants from 552 institutions, highlighting the growing interest in AI technology within the community.

This initiative was developed with support from the Ministry of Science and ICT and the IITP, through projects aimed at optimizing execution environments for neural network applications and creating generative AI support systems. As ETRI continues to innovate, the tools and frameworks released are poised to transform how industries incorporate AI into their operations, fostering greater efficiency and productivity across the board.