Research‎ > ‎

Machine Learning

1. Medical Data Analysis
Healthcare Artificial Intelligence Market is growing fast and Medical imaging and diagnosis should witness more than 40% growth to surpass USD 2.5 billion by 2024. 
Behind this market growth, there is the rapid development of Deep Learning technology. Especially, Convolutional Neural Network (CNN) is mainly used for classification
and segmentation of medical image data (MRI, CT, IVUS, etc.)

The purpose of Classification is to classify the input at the image-level while the purpose of Segmentation is to classify the input at the pixel-level. Since Medical 
Image data differs from general dataset such as ImageNet, Cifar10, UECFood256, it requires Novel CNN model considering Characteristics of Medical Image data.

Therefore, we design an optimized machine learning algorithm that considers the characteristics of medical data. We are also working in collaboration with major
hospitals including Asan Medical Center and Korea University Medical Center

2. Time Series Prediction
Time series is series of data points indexed in time order usually successively equal spaced. Time series prediction is a high dimensional
regression problem 
and there are two major groups of regression approaches: 1. Regression (Ridge, Nerual Network, Linear), 2. Probabilistic (Gaussian, Laplace)

Recurrent Neural Network (RNN) has good capabilities for time series prediction because it is robust to noise, is nonlinear, and has multivariate inputs.
However, there 
is the 'vanishing/exploding gradient' problem in RNN. Therefore, we design novel Echo State Network which only trains the hidden-output connections.