Jinseok Kim
Contact: jinseok at rebellions.ai (work), jinseokim804 at gmail.com (personal)
Hello, I'm Jinseok Kim, currently a research scientist at Rebellions Inc.. We aim to provide the world with an energy-efficient domain-specific AI processor. For more information about us, please refer to our website.
Before joining Rebellions, I finished my PhD at Pohang University of Science and Technology (POSTECH), Korea, advised by Professor Jae-Joon Kim in Device and Integrated Circuit Engineering (DICE) Lab.
Experience
Research Scientist, Rebellions Inc., Korea
AI Accelerator & Systems
Feb. 2021 - Current
Research Intern, IBM T.J. Watson Research Center, USA
Mentor: Dongsoo Lee, Manager: Pong-Fei Lu
Reduced precision deep neural network training
Jun. 2016 - Sep. 2016
Education
Pohang University of Science and Technology (POSTECH), Korea
Ph.D. in Creative IT Engineering (CiTE)
Advisor: Prof. Jae-Joon Kim
Mar. 2015 - Feb. 2021
B.S. in Creative IT Engineering (CiTE)
Graduated Summa Cum Laude, GPA: 4.08/4.3
Mar. 2012 - Feb. 2015
State University of New York (SUNY) at Stony Brook, USA
Exchange Student
GPA: 4.0/4.0
Sep. 2012 - Dec. 2012
Korea Science Academy (KSA) of KAIST, Korea
Mar. 2009 - Feb. 2012
Research Interests
My research interests range from algorithm development to chip design related to neuromorphic engineering and deep learning. Previous/current research projects in which I played a major role include
Devising a novel supervised learning method for Spiking Neural Networks considering the unique characteristics of spiking neurons [Neurips2020] (Arxiv Link)
FPGA- and ASIC-level development and demonstration (link) of Spiking Neural Network acceleration hardware (named as FireFly) [Frontiers2018]
Developing environments and applications for crossbar-based neuromorphic hardware [ISLPED2018]
SRAM-based in-memory Binary Neural Network hardware design [VLSI2019]
Publications
Jinseok Kim, Kyungsu Kim, Jae-Joon Kim, "Unifying Activation- and Timing-based Learning Rules for Spiking Neural Networks", Conference on Neural Information Processing Systems (NeurIPS), Dec. 2020, (Arxiv Link)
Hyungjun Kim, Kyungsu Kim, Jinseok Kim, Jae-Joon Kim, "BinaryDuo: Reducing Gradient Mismatch in Binary Activation Network by Coupling Binary Activations", 8th International Conference on Learning Representations (ICLR), Apr. 2020.
Jinseok Kim, Jongeun Koo, Taesu Kim, Yulhwa Kim, Hyungjun Kim, Seunghyun Yoo, Jae-Joon Kim, "Area-Efficient and Variation-Tolerant In-Memory BNN Computing using 6T SRAM Array," Symposium on VLSI Circuits (SOVC), Jun. 2019
Jongeun Koo, Jinseok Kim, Sungju Ryu, Chulsoo Kim, Jae-Joon Kim, "Area-Efficient Transposable 6T SRAM for Fast Online Learning in Neuromorphic Processors," IEEE Custom Integrated Circuits Conference (CICC), Apr. 2019
Jinseok Kim, Jongeun Koo, Taesu Kim, and Jae-Joon Kim, "Efficient Synapse Memory Structure for Reconfigurable Digital Neuromorphic Hardware," Frontiers in Neuroscience, Nov. 2018
Jinseok Kim, Yulhwa Kim, Sungho Kim, and Jae-Joon Kim, "Compact Convolution Mapping on Neuromorphic Hardware using Axonal Delay," IEEE/ACM International Symposium on Low Power Electronics and Design (ISLPED), Jul. 2018 (nominated for best paper award)
Hyungjun Kim, Taesu Kim, Jinseok Kim, Jae-Joon Kim, "Deep Neural Network Optimized to Resistive Memory with Nonlinear Current-Voltage Characteristics," ACM Journal on Emerging Technologies in Computing Systems (JETC), Jun. 2018
Taesu Kim, Hyungjun Kim, Jinseok Kim, and Jae-Joon Kim, "Input Voltage Mapping Optimized for Resistive Memory-Based Deep Neural Network Hardware," IEEE Electron Device Letters, Sep. 2017
Recognition
Scholarships
NRF (National Research Foundation of Korea) Global Ph.D. Fellowship Program
"Reconfigurable Neuromorphic Hardware with On-chip Learning Capability for Deep Spiking Neural Networks"
Jul. 2016 - Dec. 2020
POSTECH-SUNY Korea IT Consilience Creative Program (ITCCP)
Mar. 2012 - Dec. 2020
Awards
Silver Prize (2nd place) in 21st Annual Samsung HumanTech Paper Award in Computer Science & Engineering Division
"Resistance Drift Tolerant Data Coding for 4-Level-Cell Phase-Change Memory"
Feb. 2015
The best student of the year in ICT Consilience Creative Program
MSIP (Ministry of Science, ICT and Future Planning), Korea
Aug. 2014