M.IN.D. Lab @ SKKU

Main research areas & tools

  • Deep learning / Machine learning

  • Statistical signal processing

  • Optimization

  • Information theory

  • High-dimensional statistics

  • Main tools: Python/NumPy, C++/CUDA, Caffe, Theano, Tensorflow, Keras

Current projects

Core learning algorithms

  1. Neural universal denoising / smoothing

    1. Developing a novel framework of combining deep learning with denoising/estimation, which are fundamentally different from the supervised learning problems.

    2. Supported by NRF Young Researcher Program (한국연구재단 신진연구)
      (Funding: 100M KRW/year, 2016.6~2019.5)

  2. Adaptive machine learning algorithm for intelligent autonomous digital companion

    1. Developing Bandit Parameter Estimation (BPE) algorithm, Multi-Task/Lifelong Reinforcement Learning (MT/LL-RL) algorithm for digital companion (DC)

    2. Supported by MSIP-IITP Flagship Project on AI (미래창조과학부 지능정보기술 플래그십 프로젝트 참여(주관:KAIST))
      (Funding: 150M KRW/year, 2016.12~2020.12)

  3. Deep Learning based knowledge augmented reasoning

    1. Developing algorithm for machine comprehension (e.g., text-based question-answering) system that can construct and learn from external knowledge database

    2. To be supported by Samsung Software R&D Center (삼성전자 소프트웨어 R&D 센터)
      (Funding: 80M KRW/year, 2017.4~2017.12)

  4. Distributed bootstrap for model selection

    1. Devising a method based on the Bag-of-Little-Boostrap (BLB) that can speed-up cross validation (CV) for Big Data

    2. Joint work with Prof. Bin Yu (UC Berkeley, Statistics)

(Big) Data science applications

  1. Non Intrusive Load Monitoring (NILM) / Energy disaggregation

    1. Disaggregating the energy usage per device from the total energy usage (time-series) data, based on convolutional neural networks (CNN).

    2. Supported by Encored Technologies,Inc. (Funding: 25M KRW/year, 2016.10~2017.12)

  2. Fast and accurate PM2.5 monitoring

    1. Applying deep learning techniques to accurately infer the PM2.5 level based on the satellite data

    2. Joing work with Prof. Yang Liu (Emory University, Environmental Health)

  3. Micro-Doppler based activity/object recognition

    1. Attempting fine-grained object recognition based on micro-Doppler radar data, based on convolutional neural networks (CNN).

    2. Joint work with Prof. Youngwook Kim (CSU Fresno, ECE)

  4. DNA sequence denoising

    1. Devising fast, flexible, and robust denoising scheme for NGS DNA sequence data

    2. Joint work with Prof. Sungroh Yoon (SNU, ECE)

  5. Medical data analyses

    1. Applying machine learning techniques to brain images/data for predictive medicine

    2. Joint work with Prof. Jae-Jin Song (SNUH) and Dr. Jangsup Moon (SNUH)

Potential future topics

  • Confidence intervals for matrix completion schemes

  • Model compression for deep learning

  • Concentration inequalities

  • Content recommendation for SNS

Past projects

  • Remote sensing of AOD

  • Web search ranking and recommendation

  • Shifting DUDE

  • Speech recognition