Socar (KRX: 403550) announced that it will present two papers on deep learning research at the Practical Machine Learning for Developing Countries (PML4DC) workshop organized by the International Conference on Learning Representations (ICLR).
The papers focus on efficient data training in resource-poor environments and demonstrate Socar's commitment to advancing AI applications in developing countries.
The first paper, "Uncovering Effectiveness of Calibration on Open Intent Classification," proposes a method to add calibration to the loss when a deep learning model learns data with cross-entropy loss.
This approach demonstrates how a deep learning model can effectively classify category sentences not included in the existing training data as separate out-of-distribution (OOD) labels in small dataset environments.
The second paper, "Simultaneous Utilization of Part-of-Speech Replacement and Feature Space Interpolation for Text Data Augmentation (PMixUp)," introduces a data augmentation technique that applies synonym replacement and feature space interpolation in situations where practical model training is difficult due to insufficient data.
This method can solve sentence classification problems with remarkable performance even when data is scarce.
Socar's research aims to improve the efficiency of platform operation and develop AI models that can best understand Socar's domain based on natural language data obtained through platform operation.
The company plans to launch an AI customer center solution based on this AI model as early as this year, using a self-developed Large Language Model (LLM) to build conversational AI.
This technology will enable quick and accurate responses to customer inquiries while operating the vehicle.
In addition, Socar's proprietary Speech-to-Text (STT) technology is expected to improve service quality by transcribing calls between customers and agents and analyzing customer queries.
Socar has already integrated AI into its operations by combining data from platforms such as apps and vehicles.
For example, in vehicle accidents, Socar uses photos, DR-GPS sensors, and video data to process vehicle damage or accident information using AI models automatically.
The company is also automating various processes in carwashing, such as dirtiness detection, wash quality verification, and carwash certification.
The presentation of this research at ICLR 2023 highlights Socar's continued commitment to leveraging data and technology to solve and streamline various problems within its platform while sharing its deep learning expertise with the global community.