[Asia Economy Reporter Hyungsoo Park] Global medical artificial intelligence (AI) solution company VUNO announced on the 7th that it will present a research paper on an algorithm that enhances the reliability of AI-generated results by enabling deep learning models to autonomously detect data they have not previously learned at the world's most prestigious AI conference (AAAI).
AAAI, established in 1979, is a conference where leading AI research institutions and representative companies from various countries gather annually to share the latest AI technologies and research achievements. Papers presented at the AAAI conference provide a new research foundation for AI scholars worldwide. The 37th 'AAAI 2023' will be held in Washington DC, USA, until the 14th.
The research was conducted to increase the reliability of AI results presented to users. The VUNO research team developed an algorithm that enables deep learning models to autonomously detect 'out-of-distribution data.' Out-of-distribution data refers to data that deep learning models have not previously learned.
When inputting images or videos of different modalities into AI medical devices trained on fundus images, the AI must provide information indicating that the analysis results based on those images may be uncertain. In fields where AI reliability and safety are critical, such as diagnostic assistance in medical areas or autonomous driving, it is necessary for AI to clearly inform users about data outside the categories it has learned.
The VUNO research team developed KIRBY, an algorithm that autonomously detects out-of-distribution data. KIRBY works by generating virtual out-of-distribution data samples, then having the model distinguish between previously learned data and newly created sample data, enabling self-learning. Through this, it was confirmed that deep learning models can identify data they have not previously learned and provide users with additional information regarding uncertainty.
VUNO is preparing to register patents for this algorithm and technology. By applying it to its key products, the company plans to significantly enhance user trust in its products. Based on various related patented technologies, including the design and construction of existing deep learning models and data collection, VUNO will continue diverse research and development efforts to advance its products.
Lee Yeha, CEO of VUNO, said, "This research achievement is significant as it demonstrates VUNO's research and development capabilities at the world's most prestigious conference amid the global discussions on artificial intelligence sparked by the recent ChatGPT craze."
She added, "The use of artificial intelligence in the medical field is moving beyond its infancy and entering a mature phase. We will apply our products to more clinical sites and continue to advance our products continuously."
© The Asia Business Daily(www.asiae.co.kr). All rights reserved.
