A Study on Meta-Learning models and Its Applications for integrated Analysis of Heterogeneous Big Data
Along with the production of big data from a variety of sources, the demand for analysis of big data from different sources is gradually increasing. However, not many cases have been suggested yet for the deep learning based analysis of heterogeneous big data. Accordingly, this study proposes a deep learning based heterogeneous big data analysis model. The goal of this research is development of meta learning models and it’s application methodology for integrated analysis of heterogeneous big data. To achieve this goal, we develop various learning models based on k-shot learning, transfer learning, generative adversarial model, and methods for enhancing learning efficiency based on reinforcement learning. In addition, we develop dataset for heterogeneous big data integrated analysis research, study heterogeneous big data analysis model based on meta-learning, and develop original technologies that can be applied to various problems such as cross-modal retrieval and multi-modal classification and recommendation.
- Ministry of Science and ICT
- Next-Generation Information Computing Development Program
- 2017.11.01 ~ 2021.06.30
<Heterogeneous Data Embedding Demo>
Development of an object recognition model available for a small accelerator by using convolutional neural networks
In the object recognition area, various models are proposed based on convolutional neural networks, and the performance is also updated every year. However, most object recognition models currently open to the public have a large number of hidden layers in terms of the structure of the neural network, and the number of nodes in each layer and the number of parameters are basically required to be many millions of operations. In this assignment, we implement the AlexNet object recognition model and study the processing such as creation, storage, and transformation of learned connection parameters. We also research a more efficient object recognition model based on the combined-product neural network and find efficient ways to reduce the number of parameters and the size of the neural network while maintaining similar accuracy. Through such research, it is expected that the deep learning based object recognition system will be applied to a small embedded system such as a mobile device.
- Industry-University cooperation research and development
- 2017.01.01 ~ 2017.12.31
Development of Intelligence FinBot service system using CBR(Case Based Reasoning)
Intelligent Financial Bot Auto Consulting Service is the first mobile messenger-based automatic consultation service in Korea. When the user inquires through a mobile messenger (eg, IBK Talk, KakaoTalk, line, etc.) using a smartphone, the financial bot system analyzes the meaning of the contents received from the customer, It is a real-time financial bots consulting service for artificial intelligence that responds automatically to customer’s smartphone. The main research subjects of this project are research, design and implementation of answer inference engine to the requirements of the consultant.
- Ministry of Trade, Industry and Energy
- 2016.06.01 ~ 2017.05.31
Developed an assistant Chatbot for customer satisfaction service(CS)
Utilizing a messenger service that is popular among users in Korea, a chat channel for providing a customer consulting service is secured. We will develop a system that can enhance the convenience and service quality of customers and agents by combining new channels and intelligent chat robots. It is possible to utilize data secured by channels such as telephone, online, and mobile to acquire knowledge dictionary specialized for customer consultation service, and to utilize artificial intelligence learning method as soon as possible, Develop a chat bot.
- Core University of Software(Mnistry of Science, ICT and Future Planning), CNTtech(c)
- I.EAP Industry-University Cooperation
- 2016.06.01 ~ 2017.02.28
A study on semi-supervised learning model for big data and its application
Although there have been many researches on supervised learning and unsupervised learning methods in existing studies on the analysis of data streams, there are not many cases in which the data are actually commercialized and distributed. In particular, classification is performed while adapting to the concept change without labeled data There were few companies that commercialized a model of semi-supervised learning that could be done. Using the classification method of data stream developed in this study, it is possible to solve the problem of concept variation in which characteristics change with time. Also, reliance on labeled training data is reduced and can be practically applied to commercial systems.
- Ministry of Education
- Science & Engineering Individual Basic Research Support Project
- 2015.11.01 ~ 2018.10.30
AR & Recommender System
Development of Augmented Reality Technology for Mobile Devices to Promote Advertising
Augmented Reality(AR) is a technique of superimposing virtual objects on the real world seen by the user. It is also referred to as mixed reality (MR) because the virtual worlds with additional information in real time are combined into one image. By allowing users to immerse themselves in a virtual environment, they can see the real environment and provide better realism and additional information. For example, if you shine a smartphone camera around you, information such as the location of a nearby shop, phone number, etc. will be displayed as a stereoscopic image. Remote medical diagnosis, broadcasting, architectural design, and manufacturing process management. Recently, smart phones have become widely used and commercialized. Game and mobile solutions industry, and education. In this research and development project, we implemented augmented reality system through synthesis of 3D dance avatar and real world by using smart phone as terminal device, introduced personalized recommendation system, and integrated real time mobile Developed an advertising platform.
- Small and Medium Business Administration
- Industry-University-Research Laboratory Joint Technology Development Project
- 2010.6.1 ~ 2011.5.31
Self-led E-training service technology development for car maintenance training
Expert system is a system designed to accumulate experts’ expertise, experience and know-how in the computer so that they can have the same or better problem solving ability as the expert. The expert system firstly describes the characteristics of the problem, identifies the basic concepts expressing knowledge, and determines the structure and performance of the knowledge through the process of determining the structure for organizing knowledge. Expert systems are applied to fields requiring human intellectual abilities, including medical diagnostics, equipment failure diagnostics, stock investment decisions, production scheduling, automotive fault diagnosis, effective job placement, material procurement schedules, and management planning is. In this project, we develop virtual reality based training system for basic principles training and operation principle training for standard car model and develop self – directed cooperative learning training model based on logical model.
- Ministry of Knowledge Economy
- Knowledge Service USN Industry Source Technology Development Project
- 2011.6.1 ~ 2014.5.31