Minkyu Choi

Research interests
  • Deep learning, Reinforcement learning and Robotics.
  • Cognitive Neuro-Robotics, brain-inspired AI.


Research Experience
(08.2018 - Present) Purdue University, West Lafayette, USA
    Research Assistant, Professor Zhongmin Liu

(08.2017 - 07.2018) Okinawa Institute of Science and Technology, Japan
    Research staff
  • Variational Inference for predictive coding vision
  • Goal-directed action planning with predictive coding

(01.2015 - 08.2017) KAIST - Cognitive Neuro-Robotics Lab, Daejeon, Korea
    Research Assistant, Professor Jun Tani
  • Robotics system for goal-directed skilled behavior; Developed an affordance-based robot agent under various task environment with repeated training by using deep learning schemes. With a robotic arm, planning, navigating and grasping tasks are investigated.
  • Predictive coding-type vision system for robots; Proposed a novel predictive coding-type deep learning system for robots. This model is able to recognize as well as predict current input streams simultaneously. Brain-inspired structures are widely implemented (top-down and bottom-up connection, multi spatio-temporal hierarchy, predictive coding scheme). Also, developing memory in the form of attractor dynamics is scrutinized and compared to human brains.
  • Visuo-motor learning for robot; Developed a robotic system capable of learning both vison and motor signals. The system learned to map different types of signal to each other.
  • Vision system for glasses-type wearable devices; Developed a system interpreting on-sight visual scene by considering contextual information. The system is implemented in the glass-type wearable devices to assist people to make better decisions.
  • Multi-task Reinforcement learning with Parametric Bias; By introducing Parametric Bias (PB) to the deep-reinforcement leraning, agent showed improved performance for multi-task.
  • Attention based image recognition; Developed attention based image recognition deep-learning model. In order to reduce computational cost for robots with low-cost hardware environment, attention based approach is adopted.  
(03.2014 - 12.2014) Yonsei University - Digital Signal Processing Lab, Seoul, Korea
    Research Assistant, Professor Hong-gu Kang
  • Autonomous driving system for mobile robots; Develop an autonomous agent actively interpreting signs on the streets. The agent does not have to have full maps to navigate. It is designed to guide itself by interpreting clues from visual signs and symbols on the streets.

Teaching Experience
  • (Spring, 2017) EE202 Signal and System, KAIST
    • Lecturer 
  • (Fall, 2016) EE817 Deep Learning and Dynamic Neural Network Models, KAIST
    • Teaching Assistant – Supervised class projects and experiments
  • (Fall, 2016) International Student Supporting Committee, KAIST
    • Teaching Assistant – Academic support for international students
  • (Spring, 2016) EE202 Signal and System, KAIST
    • Lecturer 
  • (Summer, 2013, 2014) Robotics Camp for Youth
    • Organizer & Lecturer – Organize and lead robotics camp for youth. Introduce basic robotics

  • Journal publication
    • Choi, M., & Tani, J. (2018). Predictive Coding for Dynamic Visual Processing: Development of Functional Hierarchy in a Multiple Spatiotemporal Scales RNN Model. Neural computation, 30(1), 237-270. [PDF]
    • Hwang, J., Kim, J., Ahmadi, A., Choi, M. and Tani, J., 2018. Dealing With Large-Scale Spatio-Temporal Patterns in Imitative Interaction Between a Robot and a Human by Using the Predictive Coding Framework. IEEE Transactions on Systems, Man, and Cybernetics: Systems. [PDF]
  • Conference publications
    • Choi, M., Matsumoto, T., Jung, M. and Tani, J., 2018. Generating goal-directed visuomotor plans based on learning using a predictive coding type deep visuomotor recurrent neural network model. arXiv preprint arXiv:1803.02578. [PDF]
    • Hwang, J., Kim, J., Ahmadi, A., Choi, M. & Tani, J. (2017, March). Predictive Coding-based Deep Dynamic Neural Network for Visuomotor Learning. In 7th Joint IEEE International Conference on Development and Learning and on Epigenetic Robotics, In press. [PDF]
    • Choi, M. & Tani, J. Predictive coding for dynamic vision: Development of functional hierarchy in a multiple spatio-temporal scales RNN model.” 2017 International Joint Conference on Neural Networks (IJCNN) (2017): 657-664. Oral presentation. [PDF]
    • Hwang, J., Jung, M., Madapana, N., Kim, J., Choi, M. & Tani, J. (2015, November). Achieving" synergy" in cognitive behavior of humanoids via deep learning of dynamic visuo-motor-attentional coordination. In Humanoid Robots (Humanoids), 2015 IEEE-RAS 15th International Conference on (pp. 817-824). IEEE. [PDF]

  • Ross Fellowship, Purdue University
  • National full scholarship, KAIST
  • National Science and Engineering scholarship, Yonsei University
  • Honor scholarship for Academic Excellence, Yonsei University 

  • 2014, Capstone design competition, Silver Prize.
    • Designed an autonomous agent navigating itself without maps by interpreting signs and symbols for human.
  • 2014, Awards for Academic Excellence.
  • 2010, Samsung Robotics Competition, 3rd Prize. 
    • Competition for designing innovative robots for public health. Designed robot for cleaning inside public water pipe. 
  • 2010, Award for Academic Excellence.

Graduate Courses taken (MS, KAIST)
  • Neural Networks (EE538)
  • Neuro-Robotics (EE837)
  • Deep Learning and Dynamic Neural Network Models (EE817)
  • Statistical Learning Theory (EE531)
  • Theory of Brain Function (BiS527)
  • Reinforcement Learning (IE801)

  • OS: Windows, Linux
  • Programming Languages: C/C++, Python, Java
  • Tools and programs: Matlab, OpenCV, Tensorflow, Cuda
  • Robots: iCub, Nao, Dimitri (humanoid robot built in Lab), Torobo (Tokyo robotics)
 iCub    Nao