Project 1 - Deep learning based wafer bin map detection and map fail pattern detection (Samsung Electronics) - 2018.04 ~ 2018.11

Daily WBM clustering and pattern generation rate detection methodologies based on specific dates

WBM has different patterns depending on various products, process routes, etc., and needs to search for specific pattern change of WBM per day / process unit

The fact that the pattern of WBM changed much compared to the past in the process schedule and materials.

This study seeks to detect when different patterns of WBM are generated for a day in similar process situations and to provide engineer with priority information of process profiling.

Analysis of daily WBM is conducted by using Convolution Auto-Encoder(CAE) method that can extract variables for WBM abnormal pattern of each product.

  Project 2 - Neuroimage analysis of major depressive disorder in the adolescents based on machine learning/deep learning (Korea Hospital) - 2017.08 ~ 2018.02

The primary objective is to train a high accuracy predictive classification that could discriminate scans from adolescent patients with MDD from scans from healthy controls.

We demonstrate whether it is possible to reliably train a predictive classification with a high accuracy even in the first-onset drug-naïve adolescent MDD just using structural MRI.

This is the first study to estimate prediction models for first-onset episode, drug-naïve adolescent MDD by evaluating the potential of structural MRI measures as biomarkers and multiple machine learning methods

  Project 3 - Development of machine learning media streaming optimization system using QoE information (INI Soft) - 2017.05 ~ 2019.02

In order to efficiently serve large-capacity media contents such as Videos on the Internet and mobile, we developed a machine learning-based algorithm and library that automatically arranges content on CDN (Content Delivery Network).

Development of QoE (Quality of Experience) Report collection technology for users' use of content.

Development of Dynamic Content Placement and CDN Selection Algorithm for Performance/Price Optimization under Multi-CDN.

Development of QoE (Quality of Experience) Report System Technology.

Multi-CDN performance/price optimization algorithm.

  Project 4 - Knowledge-based Knowledge Economy (NC Soft) - 2017.05 ~ 2018.03

Purpose of meaningful knowledge extraction from stock price data (structured) and news articles (unstructured).

Extract and visualize key events based on network analysis through news articles.

Extract news article event pattern through representation of news article.

Based on the similarity of representation of articles in each period.

Similar past searches and stock trends based on Similarity/Distance.

  Project 5 - Developing Knowledge Extraction Using News articles and Stock Price (NC Soft) - 2016.10 ~ 2017.03

Extract meaningful information based on structured/unstructured data in financial sector.

Development of a methodology to interpret formal data (stock prices) from Text (news articles).

Keyword extraction and generalization that implicitly represent the contents of the article from news articles.

Extraction of links between financial and economic articles and stock prices using machine learning.

Interpretation of the relationship between news articles and stock price trends in three aspects (real-time, leading, and trailing).

  Project 6 - Opinion Generation Technology Development (NC Soft) - 2015.05 ~ 2015.07

Developing a query-response methodology that provides appropriate answers based on formal/unstructured data in a domain for a human-to-conversational query.

Convert conversational queries into searchable query sets.

Extract necessary information from both structured data and unstructured data for attributes corresponding to the query set.

Combine the extracted information to express the degree of positive/negative of the evaluation factor.