Kimin Lee

I'm a postdoctoral fellow at UC Berkeley, working with Pieter Abbeel. Before this, I received a PhD at Korea Advanced Institute of Science and Technology (KAIST), where I was supervised by Jinwoo Shin. During PhD, I also interned and collaborated closely with Honglak Lee at University of Michigan.

  • Contact: kiminlee at berkeley dot edu // pokaxpoka at gmail dot com

Updates

  • (2020/09): I am co-organizing NeurIPS 2020 Deep RL Workshop

  • (2020/06): I receive the best doctoral thesis award from KAIST EE

  • (2019/11): I'll be a postdoctoral fellow at the UC Berkeley working with Pieter Abbeel

  • (2019/03): I'll visit the University of Michigan (working with Honglak Lee)

  • (2018/05): I'll visit the UC Berkeley (working with Dawn Song and Bo Li)

Publications (C: conference / J: journal / P: preprint / *: equal contributions)

2021

[P9] Improving Computational Efficiency in Visual Reinforcement Learning via Stored Embeddings [pdf][code]

  • Lili Chen, Kimin Lee, Aravind Srinivas, Pieter Abbeel

  • Arxiv preprint 2021

  • Previous version [pdf] [video] in NeurIPS Workshop on Deep RL 2020

[P8] State Entropy Maximization with Random Encoders for Efficient Exploration [pdf][code][site]

    • Younggyo Seo*, Lili Chen*, Jinwoo Shin, Honglak Lee, Pieter Abbeel, Kimin Lee

    • ICLR Workshop on Self-Supervision for RL 2021

[C17] Learning to Sample with Local and Global Contexts in Experience Replay Buffer [pdf][code]

[C16] MASKER: Masked Keyword Regularization for Reliable Text Classification [pdf][code]

2020

[P7] Addressing Distribution Shift in Online Reinforcement Learning with Offline Datasets [pdf] [video]

  • Seunghyun Lee, Younggyo Seo, Kimin Lee, Pieter Abbeel, Jinwoo Shin

  • NeurIPS Workshop on Offline RL 2020 as Oral presentation

  • NeurIPS Workshop on Deep RL 2020

[P6] R-LAtte: Visual Control via Deep Reinforcement Learning with Attention Network [pdf] [video]

  • Mandi Zhao, Qiyang Li, Aravind Srinivas, Ignasi Clavera, Kimin Lee, Pieter Abbeel

  • NeurIPS Workshop on Deep RL 2020

[P5] Decoupling Representation Learning from Reinforcement Learning [pdf][code]

    • Adam Stooke, Kimin Lee, Pieter Abbeel, Michael Laskin

    • NeurIPS Workshop on Deep RL 2020

[P4] SUNRISE: A Simple Unified Framework for Ensemble Learning in Deep Reinforcement Learning [pdf][code]

    • Kimin Lee, Michael Laskin, Aravind Srinivas, Pieter Abbeel

    • NeurIPS Workshop on Deep RL 2020

[P3] Reinforcement Learning for Sparse-Reward Object-Interaction Tasks in a First-person Simulated 3D Environments [pdf][podcast]

    • Wilka Carvalho, Anthony Liang, Kimin Lee, Sungryull Sohn, Honglak Lee, Richard L Lewis, Satinder Singh

    • NeurIPS Workshop on Deep RL 2020

    • Previous version [pdf] in ICML Workshop on OOL 2020

[C15] Trajectory-wise Multiple Choice Learning for Dynamics Generalization in Reinforcement Learning [pdf][code][site]

[C14] Reinforcement Learning with Augmented Data [pdf][code][blog] [media]

[C13] Context-aware Dynamics Model for Generalization in Model-Based Reinforcement Learning [pdf][code][site]

[C12] Regularizing Class-wise Predictions via Self-knowledge Distillation [pdf] [code]

[C11] Network Randomization: A Simple Technique for Generalization in Deep Reinforcement Learning [pdf] [code]

[P2] Dynamics Generalization via Information Bottleneck in Deep Reinforcement Learning [pdf]

    • Xingyu Lu, Kimin Lee, Pieter Abbeel, Stas Tiomkin

    • Arxiv preprint 2020

2019

[C10] Overcoming Catastrophic Forgetting with Unlabeled Data in the Wild [pdf][code]

[C9] Robust Inference via Generative Classifiers for Handling Noisy Labels [pdf][code]

[C8] Using Pre-Training Can Improve Model Robustness and Uncertainty [pdf] [code]

[J1] Dynamic Control for On-demand Interference-managed WLAN Infrastructures [pdf]

    • Seokhyun Kim, Kimin Lee, Yeonkeun Kim, Jinwoo Shin, Seungwon Shin, Song Chong

    • IEEE/ACM Transactions on Networking (TON) 2019

2018

[C7] A Simple Unified Framework for Detecting Out-of-Distribution Samples and Adversarial Attacks [pdf] [code]

[C6] Learning to Specialize with Knowledge Distillation for Visual Question Answering [pdf] [code]

[C5] Hierarchical Novelty Detection for Visual Object Recognition [pdf] [code]

[C4] Training Confidence-calibrated Classifiers for Detecting Out-of-Distribution Samples [pdf] [code]

2017

[C3] Confident Multiple Choice Learning [pdf] [code]

[P1] Simplified Stochastic Feedforward Neural Networks [pdf]

    • Kimin Lee, Jaehyung Kim, Song Chong, Jinwoo Shin

    • Arxiv preprint 2017

2016

[C2] TravelMiner: On the Benefit of Path-based Mobility Prediction

[C1] Just-in-time WLANs: On-demand Interference-managed WLAN Infrastructures [pdf] [slides]

Education

  • 2015.02-2020.02: Doctor of Philosophy (Ph.D.), School of Electrical Engineering (Machine/deep learning), KAIST (adviser: Jinwoo Shin)

  • 2013.02-2015.02: Master of Science (MS), School of Electrical Engineering (Wireless communication networks), KAIST (adviser: Song Chong and Jinwoo Shin)

  • 2009.02-2013.02: Bachelor of Engineering (BS), School of Electrical Engineering, KAIST

Presented Talks

  • "Scaling deep reinforcement learning to challenging domains", Seoul National University, Seoul, Korea, 2021 (virtual)

  • "Scaling deep reinforcement learning to challenging domains", KAIST, Daejeon, Korea, 2021 (virtual)

  • "Ensemble method for reinforcement learning", BAIR seminar, USA, 2020 (virtual)

  • "Data-efficient reinforcement learning from high-dimensional inputs ", Prof. Joseph J. Lim's Group, USC, USA, 2020 (virtual)

  • "Generalization/robustness in deep reinforcement learning ", Prof. Bo Li's Group, UIUC, USA, 2020 (virtual)

  • "Recent progress and challenges in deep reinforcement learning ", POSTECH, Pohang, Korea, 2020 (virtual)

  • "A simple randomization technique for generalization in deep reinforcement learning", NeurIPS Workshop on DRL, Vancouver, Canada, 2019

  • "Robust inference via generative classifiers for handling noisy labels", Samsung Research, Seoul, Korea, 2019

  • "Generative classifier for detecting out-of-distributions and handling noisy labels", Samsung DS, Suwon, Korea, 2019

  • "Toward overcoming out-of-distribution examples in deep learning", Prof. Pieter Abbeel's Group, Berkeley, CA, USA, 2019

  • "Toward overcoming out-of-distribution examples in deep learning", Prof. Dawn Song's Group, Berkeley, CA, USA, 2019

  • "Robust inference via generative classifiers for handling noisy labels", ICML, Long Beach, USA, 2019

  • "Generative classifier for detecting abnormal samples and handling noisy labels", Samsung Advanced Institute of Technology, Suwon, Korea, 2019

  • "Improving the reliability and robustness of deep learning", LG CNS AI, Seoul, Korea, 2019

  • "A simple unified framework for detecting out-of-distribution samples and adversarial attacks", NeurIPS, Montreal, Canada, 2018

  • "Generative classifier for obtaining well-calibrated predictive uncertainty and developing a robust inference method", Google DeepMind, Mountain View, CA, USA, 2018

  • "Training confidence-calibrated classifiers for detecting out-of-distribution samples", Samsung Advanced Institute of Technology, Suwon, Korea, 2018

  • "Training confidence-calibrated classifiers for detecting out-of-Distribution samples", NAVER Clova AI, Seoul, Korea, 2018

  • "Confident multiple choice learning", SK telecom, Seoul, Korea, 2017

  • "Confident multiple choice learning", ICML, Sydney, Australia, 2017

  • "Just-in-time WLANs: on-demand interference-managed WLAN infrastructures", INFOCOM, San Francisco, USA, 2016

Academic Activities

  • Conference reviewer

    • NeurIPS, ICML, ICLR, AAAI, ICRA, UAI, IJCAI

  • Journal reviewer

    • Journal of Machine Learning Research, IEEE Transactions on Pattern Analysis and Machine Intelligence, IEEE/ACM Transactions on Networking

  • Workshop organizer

    • Deep Reinforcement Learning Workshop, NeurIPS 2020

  • Admissions committee

    • UC Berkeley EECS, 2020

Work Experience

  • 2020.02 - current: Postdoctoral fellow at UC Berkeley (working with Pieter Abbeel)

  • 2019.04 - 2019.09: Visiting student at University of Michigan (working with Honglak Lee)

  • 2018.06 - 2018.10: Visiting student at UC Berkeley (working with Dawn Song and Bo Li)

Mentoring

  • Lili Chen (BS student at UC Berkeley)

  • Seunghyun Lee (MS student at KAIST)

  • Younggyo Seo (PhD student at KAIST)

  • Mandi Zhao (BS student at UC Berkeley)

Selected Honor

  • Best doctoral thesis award from KAIST EE

  • Excellent research achievement award from Samsung Advanced Institute of Technology