Archives: 연구 > 연구성과

Professor Dong Eui Chang of our department has published a book on deep learning titled “Deep Neural Networks in a Mathematical Framework” with Anthony L. Caterni of Oxford University. (Springer; 2018) Detailed information about the book is on below. You can download eBooks through the link https://doi.org/10.1007/978-3-319-75304-1, and bound books are available in online/offline bookstores.

Title: Deep Neural Networks in a Mathematical Framework

Authors: Anthony L. Caterini and Dong Eui Chang

Publisher: Springer; 2018

ISBN 978-3-319-75303-4

ISBN 978-3-319-75304-1 (eBook)

https://doi.org/10.1007/978-3-319-75304-1

Book cover and Front Matter: in attachment

 

Summary:

This book describes how to build a rigorous end-to-end mathematical framework for deep neural networks. The authors provide tools to represent and describe neural networks, casting previous results in the field in a more natural light. In particular, the authors derive gradient descent algorithms in a unified way for several neural network structures, including multilayer perceptrons, convolutional neural networks, deep autoencoders and recurrent neural networks. Furthermore, the authors developed framework is both more concise and mathematically intuitive than previous representations of neural networks.

This book is one-step towards unlocking the black box of Deep Learning. The authors believe that this framework will help catalyze further discoveries regarding the mathematical properties of neural networks. This book is accessible not only to researchers, professionals and students working and studying in the field of deep learning, but also to those outside of the neutral network community.

Professor Kyung Cheol Choi and Ph.D candidate Yongmin Jeon have successfully developed OLED Wearable Light therapy Patch with Professor Kyoung-Chan Park and Hye-Ryung Choi from Department of Dermatology, Seoul National University Bundang Hospital (SNUBH).

Phototherapy is a treatment that promotes the biochemical reaction of the human body by irradiating light. It is widely used to heal wounds through LED or laser devices installed in hospitals. However, conventional phototherapy equipments are not flexible, also it is difficult to uniformly irradiate light, and there is a problem that heat is generated. For these reasons, even if patients wanted to increase the effectiveness of the treatment, there was a limit that can not be attached to the human body.

The phototherapy patch developed by Professor Kyung Cheol Choi’s team is light and flexible, allowing you to maintain high-efficiency treatment while adhering to your skin in daily life. The components OLED, battery, heat sink and patch are designed in the form of thin film, less than 1mm thick and less than 1g in weight. It also operates for a long period of more than 300 hours and can be driven in a bend radius of 20 mm, so it can be attached to various body parts. It operates at less than 42°C, eliminates the risk of low-temperature burns, and it has been verified to be safe by the International Organization for Standardization (ISO) standards, and has an excellent healing effect that improves cell proliferation by 58% and cell migration by 46%.

This research was supported by the Ministry of Science and Technology, Ministry of Information and Communication / Korea Research Foundation Basic Research Project (ERC : Attachable Photo Therapeutics Center for e-Healthcare)

Published in the International Journal of Advanced Materials Technologies (March 8). More information can be found in the following papers and press releases.

<Papers>

Title: A Wearable Photobiomodulation Patch Using a Flexible Red-Wavelength OLED and Its in Vitro Differential Cell Proliferation Effects

Link: http://onlinelibrary.wiley.com/doi/10.1002/admt.201700391/full

<Press releases video link>

[MBC] ▶ OLED adhesive plaster 

https://www.youtube.com/watch?v=ePbSyaJhcMc 

 [Channel A] ▶ ‘Laser treatment’ as a bandage … Heal the wound with light

https://www.youtube.com/watch?v=1yXweMt3uX4

[Yonhapnews TV] ▶ [HOT NEWS] OLED bandages. Light treatment quickly heals hurt 

https://www.youtube.com/watch?v=Z7kR3_Hvdlo

 

Figure 1. Driving picture of wearable OLED patch attached to skin

 

Figure 2. Wound healing effect of wearable OLED patch attached to skin

Paper of Ph.D. candidate Seong Tae Kim, Hakmin Lee, Hak Gu Kim (Advisor: Prof. Yong Man Ro) was selected as Robert F. Wagner All-Conference Best Student Paper Award Finalist (Field of Computer-Aided Diagnosis Conference).

The SPIE Medical Imaging Conference is one of the world’s largest conferences on medical imaging. The selected paper is “ICADx: Interpretable computer aided diagnosis of breast masses” as a result of “Interpretable Deep Learning” study.

 

Conference: SPIE Medical Imaging (http://spie.org/conferences-and-exhibitions/medical-imaging&nbsp😉

Date: 2018.02.10 – 2018.02.15

Award: Robert F. Wagner All-Conference Best Student Paper Award Finalist (First place in Computer-Aided Diagnosis Conference)

Place: Marriott Marquis Houston, Houston, Texas, United States

Title of the paper: ICADx: interpretable computer aided diagnosis of breast masses

Authors: Seong Tae Kim, Hakmin Lee, Hak Gu Kim, and Yong Man Ro

Master candidate Sung-Jun Yoon and Ph.D. candidate Yongwoo Kim (Advisor: Prof. Munchurl Kim) received Gold Award (Selected 2 first prize out of 280 published papers) at the “Workshop on Image Processing and Image Understanding (IPIU 2018)” conference held once a year as an academic event in the field of computer vision and image processing.

-Name of the Academic Event: Workshop on Image Processing and Image Understanding (IPIU 2018)

-Event Period: February 7th (Wed), 2018 – February 9th (Fri), 2018

-Place of Event: Maison Glad Jeju Hotel

-Title of Paper: A study on frame rate improvement using hierarchical convolution neural network

-Authors: Sung-Jun Yoon, Yongwoo Kim, Munchurl Kim

Introduction:

In this paper, we propose a frame rate enhancement method using hierarchical convolution neural network. The convolutional neural network is constructed hierarchically, and the convolution operation is performed at the optimal position adaptively, so that the performance is robust against the interpolation of the fast moving object.

As a result, compared with the existing ICCV 2017 “Video Frame Interpolation via Adaptive Separable Convolution” algorithm, the improvement was 0.41dB in terms of peak signal-to-noise ratio (PSNR)

Prof. Sung-yool Choi and Prof. Sang-Hee Park (Dept. of Materials Science and Engineering)’s co-work was published as a front cover paper on January 10 in Advanced Functional Materials.

The subject of this paper is ‘Memristive Logic-in-Memory Integrated Circuits for Energy-Efficient Flexible Electronics’. The Main authors are Ph.D candidate Jang Byeong-cheol (co-author), Ph.D candidate Nam Yoon-yong (co-author), Prof. Sang Hee Park (Coressponding author), and Prof. Sung-yool Choi (Coreesponding author).

 

Journal : Advanced Functional Materials

Title : Memristive Logic-in-Memory Integrated Circuits for Energy-Efficient Flexible Electronics

Authors : Ph.D candidate Jang Byeong-cheol ,  Ph.D candidate Nam Yoon-yong (Dept. of Materials science & Engineering)

Link : http://onlinelibrary.wiley.com/doi/10.1002/adfm.201704725/full

 

 

Professor Jae-Woong Jeong and the Washington University School of Medicine in St. Louis. Northwestern University’s co-work was published in the Front Cover in Small issue of January 25.

The subject of this paper is ‘Ultra small wireless neural implant that can deliver drug and light stimulation to the brain without battery’. The authors are Raza Qazi, visiting researcher of KAIST, and Jae-Woong Jeong (Corresponding author).

 

• Paper: Small

• Title: Miniaturized, Battery-Free Optofluidic Systems with Potential for Wireless Pharmacology and Optogenetics

• Link: http://onlinelibrary.wiley.com/doi/10.1002/smll.201702479/full

Professor Munchurl Kim of our department developed a technology that can convert full HD video into super-high definition 4K UHD video using deep learning technology.

This technology has been implemented in hardware by Deep Convolution Neural Network (DNCC), which is a key techonology in artificial intelligence. It is expected to contribute premium UHD TV, 360 VR, and 4K IPTV in the future by developing algorithms and hardware that can create ultra high resolution 4K UHD screen in 60 frames per second in real time.

This study was lead by Yongwoo Kim and Jae-Seok Choi Ph.D. candidates and they are preparing for patent application.

In recent years, efforts have been made to apply Deep Convolutional Neural Network (DNCC) on image quality improvement research. However, Deep Convolutional Neural Network (DNCC) technology has a high computational complexity, and there is a limit to real-time conversion to ultrahigh-resolution images through small hardware because it needs large memory.

In the conventional frame-by-frame image processing method, it is necessary to use external memory such as DRAM, which causes memory bottleneck and power comsumption due to excessive external memory access when processing image data.

Professor Kim’s team developed an efficient Deep Convolution Neural Network structure that can process data in units of lines instead of frames to implement 4K UHD super resolution at 60 frames per second on small hardware without using external memory.

His team maintained a similar picture quality with only 65% of the filter parameters compared to the fast algorithm based on deep convolution neural networks and software.

This is the first example to implement 60 frames per second 4K UHD super resolution by using hardware.

Professor Kim said, “This research is a very important example of the Deep Convolutional Neural Network that is practically applicable to ultra-high-quality image processing in small hardware. Currently, it can be applied to premium UHD TV and UHD broadcast contents, 360 VR contents, 4K IPTV services.”  

This research was carrier out with the support from the Institute for Information & communications Technology Promotion