2024 Outstanding Paper Award at the Korean Society of Heat Treatment Engineering Autumn Conference / Sejin Park (M.S. in New Materials Engineering, General Graduate School, 24)
- 24.11.01 / 이정민
Won the Outstanding Paper Award at the 2024 Korean Society of Heat Treatment Engineering Autumn Conference / Sejin Park (M.S. Program in New Materials Engineering, Graduate School of General Studies, Kookmin University)
Sejin Park, a student majoring in new materials engineering at Kookmin University's Graduate School of General Studies, (AI Materials Design Laboratory, Prof. Ki Sub Cho) won the Best Paper Award (Jung In Sang Award) at the 2024 Korean Society of Heat Treatment Engineering Fall Conference held recently.
Sejin Park presented his paper on 'Development of data classification and quantitative data extraction system based on object detection using PDF documents'. Most research papers are a mixture of text and images, and contain important information such as alloy composition, heat treatment conditions, and physical properties. However, existing PDF processing libraries are limited in their ability to accurately extract data because they do not fully understand the complex structure and context of the document.
To solve this problem, Sejin Park developed a multimodal learning system that combines a YOLO model with a large-scale language model (LLM). In the paper, the system effectively extracts important information such as composition, heat treatment conditions, γ' phase solidification temperature, and creep rupture time from texts and images through multimodal models and builds them into a structured database.
The key achievement of this research is to simultaneously analyze image and text data of PDF documents through multimodal learning, and to improve the performance of the model through data augmentation. The existing rotation and inversion augmentation techniques have the limitation that they can distort the meaning of data in graph images where the x and y axes have specific physical meanings such as time, temperature, and experimental conditions. To overcome this, Sejin Park developed a new data enhancement technique that combines multiple single images to create a set of multiple images. This method greatly improved the generalization performance of the model by securing data diversity while preserving the physical meaning of the original image. In addition, by using Prompt Engineering to precisely analyze the interconnectedness between graphs and captions, she built a system that can more accurately classify and extract quantitative data.
By presenting an automated system that can efficiently manage and analyze large amounts of data in scientific documents and research papers, Sejin Park's research is expected to significantly speed up material development and make important contributions to various research fields.
|
This content is translated from Korean to English using the AI translation service DeepL and may contain translation errors such as jargon/pronouns. If you find any, please send your feedback to kookminpr@kookmin.ac.kr so we can correct them.
|
|
2024 Outstanding Paper Award at the Korean Society of Heat Treatment Engineering Autumn Conference / Sejin Park (M.S. in New Materials Engineering, General Graduate School, 24) |
||||
|---|---|---|---|---|
|
Won the Outstanding Paper Award at the 2024 Korean Society of Heat Treatment Engineering Autumn Conference / Sejin Park (M.S. Program in New Materials Engineering, Graduate School of General Studies, Kookmin University)
Sejin Park presented his paper on 'Development of data classification and quantitative data extraction system based on object detection using PDF documents'. Most research papers are a mixture of text and images, and contain important information such as alloy composition, heat treatment conditions, and physical properties. However, existing PDF processing libraries are limited in their ability to accurately extract data because they do not fully understand the complex structure and context of the document.
To solve this problem, Sejin Park developed a multimodal learning system that combines a YOLO model with a large-scale language model (LLM). In the paper, the system effectively extracts important information such as composition, heat treatment conditions, γ' phase solidification temperature, and creep rupture time from texts and images through multimodal models and builds them into a structured database.
The key achievement of this research is to simultaneously analyze image and text data of PDF documents through multimodal learning, and to improve the performance of the model through data augmentation. The existing rotation and inversion augmentation techniques have the limitation that they can distort the meaning of data in graph images where the x and y axes have specific physical meanings such as time, temperature, and experimental conditions. To overcome this, Sejin Park developed a new data enhancement technique that combines multiple single images to create a set of multiple images. This method greatly improved the generalization performance of the model by securing data diversity while preserving the physical meaning of the original image. In addition, by using Prompt Engineering to precisely analyze the interconnectedness between graphs and captions, she built a system that can more accurately classify and extract quantitative data.
By presenting an automated system that can efficiently manage and analyze large amounts of data in scientific documents and research papers, Sejin Park's research is expected to significantly speed up material development and make important contributions to various research fields.
|
||||
