Associate Professor

School of Computing
Graduate School of AI
KAIST

I’m an Associate Professor in the School of Computing at KAIST and also an Assistant Professor of Computer Science at Rutgers University, affiliated with the Rutgers Center for Cognitive Science. I lead the Agent Machine Learning Lab at KAIST and Rutgers. My research interest is to develop machine learning algorithms to make human-like general-purpose agents such as robots. In pursuing this, I’m particularly interested in learning representations, world models, and policies in interactive, self-supervised, and structured ways (e.g., causal, compositional, temporal, and hierarchical) with the tools of deep learning, reinforcement learning, and probabilistic learning. I also enjoy being inspired by Cognitive Science to discover novel problems and inductive biases. I received my Ph.D. at the University of California, Irvine on the study of scalable approximate Bayesian inference under the supervision of Prof. Max Welling. I did my postdoc working on deep learning at MILA under Prof. Yoshua Bengio.

Email: sjn.[last_name] at gmail
Address: E3-1 Rm:3435, KAIST, 291 Daehak-ro, Yuseong-gu, Daejeon, South Korea

Research Interest

• Reinforcement learning, representation learning, world models
• Applications: robot learning, virtual agent learning (game AI)

Openings

• Undergraduate Internship
• I’m looking for highly motivated students in all degree programs, Ph.D., M.S., and Undergraduate. If you’re interested in researching on machine learning for intelligent agents, please send me an email with your CV and transcript.

News

• Co-organizing ICLR 22 Workshop on the Elements of Reasoning: Objects, Structure and Causality.
• “ROOTS: Object-Centric Representation and Rendering for 3D Scenes” is accepted in the Journal of Machine Learning Research
• Two papers are accepted in ICML 2021
• Invited Talk at DeepMind, Jan 2021
• Our paper “Generative Neurosymbolic Machines” is accepted in NeurIPS 2020 as a spotlight!
• Invited Speaker for NeurIPS 2020 Workshop on Object Representations for Learning and Reasoning
• Co-organizing ICML 2020 Workshop on Object-Oriented Learning (https://oolworkshop.github.io/)
• Teaching in Fall 2020: CS 444: Deep Learning
• 2 papers accepted in ICLR 2020
• Teaching CS 536: Machine Learning in Spring 2020
• 3 papers accepted in NeurIPS 2019 including one spotlight paper

Media

• Mar/24/22 세계 권위 AI 학회 ‘뉴립스’에 한국인 연구자 4명 조직위원 선임
• Oct/26/21 Rutgers University’s AI Researchers Propose A Slot-Based Autoencoder Architecture, Called SLot Attention TransformEr (SLATE)

Service

• Organizing Committee - NeurIPS 2022 (Workshop Chair)
• Area Chair - NeurIPS(2021, 2022), ICML(2021,2022), AAAI(2021,2022)
• Reviewers - NeurIPS (2015-2020), ICML(2015-2020), ICLR(2015-2022), AISTATS(2015 - 2020), AAAI(2015 - 2020)

Publications / Google Scholar

2022

DreamerPro: Reconstruction-Free Model-Based Reinforcement Learning with Prototypical Representations
Fei Deng, Ingook Jang, Sungjin Ahn
ICML 22 [arXiv]

Illiterate DALLE Learns to Compose
Gautam Singh, Fei Deng, Sungjin Ahn
ICLR 22 [pdf] [project] [code]

2021

Structured World Belief for Reinforcement Learning in POMDP
Gautam Signh, Skand Peri, Junghyun Kim, Hyunseok Kim, Sungjin Ahn
ICML 21 [pdf]

Generative Video Transformer: Can Objects be the Words?
Yi-Fu Wu, Jaesik Yoon, Sungjin Ahn
ICML 21 [pdf]

Generative Scene Graph Networks
F. Deng, Z. Zhi, D. Lee, S. Ahn
ICLR 21 [pdf]

ROOTS: Object-Centric Representation and Rendering of 3D Scenes
{C. Chen, F. Deng}, S. Ahn
JMLR 21, [pdf] [project]

DreamerPro: Reconstruction-Free Model-Based Reinforcement Learning with Prototypical Representations
Fei Deng, Ingook Jang, Sungjin Ahn
NeurIPS 21 Deep RL Workshop [arXiv]

TransDreamer: Reinforcement Learning with Transformer World Models
Chang Chen, Jaesik Yoon, Yi-Fu Wu, Sungjin Ahn
NeurIPS 21 Deep RL Workshop [arXiv]

Learning Representations for Zero-Shot Image Generation without Text
Gautam Singh, Fei Deng, Sungjin Ahn
NeurIPS 21 Workshop on Controllable Generative Modeling in Language and Vision [pdf]

2020

Generative Neurosymbolic Machines
J. Jiang and S. Ahn
NeurIPS 20, [arxiv] [code] Spotlight (top 4% = 395/9454 submissions)

Improving Generative Imagination in Object-Centric World Models
Z. Lin, Y. Wu, S. Peri, B. Fu, J. Jiang, S. Ahn
ICML 20 [pdf] [project] [code]

Robustifying Sequential Neural Processes
J. Yoon, G. Singh, and S. Ahn
ICML 20 [arxiv]

SCALOR: Generative World Models with Scalable Object Representations
{J. Jiang, S. Janghorbani}, G. Melo, and S. Ahn
ICLR-20 [arxiv] [project] [code]

SPACE: Unsupervised Object-Oriented Scene Representation via Spatial Attention and Decomposition
{Z. Lin, Y. Wu, S. Peri}, W. Sun, G. Singh, F. Deng, J. Jiang, S. Ahn
ICLR 20 [pdf] [project] [code]

Hierarchical Decomposition and Generation of Scenes with Compositional Objects
F. Deng, Z. Zhi, S. Ahn
ICML 20, Workshop on Object-Oriented Learning Spotlight [pdf]

Generating Stochastic Object Dynamics in Scenes
Z. Lin, Y. Wu, S. Peri, B. Fu, J. Jiang, and S. Ahn
ICML 20, Workshop on Object-Oriented Learning [pdf]

2019

Sequential Neural Processes
{G. Singh, J. Yoon}, Y. Sohn, and S. Ahn
NeurIPS 19, Spotlight (top 2.4% = 164/6743)
[pdf] [project] [code]

Variational Temporal Abstraction
T. Kim, {S. Ahn, Y. Bengio}
NeurIPS 19

Neural Multisensory Scene Inference
J. Lim, P. Pinheiro, N. Rostamzadeh, C. Pal, and S. Ahn
NeurIPS 19

Learning Single-View 3D Reconstruction with Adversarial Training
P. Pinheiro, N. Rostamzadeh, and S. Ahn
ICCV 19, Oral (top 4.3% of all the submitted)

Generative Hierarchical Models for Parts, Objects, and Scenes
F. Deng, Z. Zhi, and S. Ahn
arXiv

Reinforced Imitation in Heterogeneous Action Space
K. Zolna, N. Rostamzadeh, Y. Bengio, {S. Ahn, P. O. Pinheiro}
arXiv

2018

Bayesian Model-Agnostic Meta-Learning
{J Yoon, T Kim}, O. Dia, S. Kim, Y. Bengio, S. Ahn
NeurIPS 18, Spotlight (top 3.5% = 168/4856)

Reinforced Imitation Learning from Observations
K. Żołna, N. Rostamzadeh, Y. Bengio, {S. Ahn, P. Pinheiro}
NeurIPS 18 Workshop on Imitation Learning and Its Challenges in Robotics

2017

Hierarchical Multiscale Recurrent Neural Networks
J. Chung, S. Ahn, Y. Bengio
ICLR 17

Denoising Criterion for Variational Auto-Encoding Framework
D. Im, S. Ahn, R. Memisevic, Y. Bengio
AAAI 17

SENA: Preserving Social Structure for Network Embedding
S. Hong, T. Chakraborty, S. Ahn, G. Husari and N. Park
ACM Hypertext and Social Media 17

2016

Pointing the Unknown Words
C. Gulcehre, S. Ahn, R. Nallapati, B. Zhou, Y. Bengio
ACL 16 Oral Presentation

Generating Factoid Questions with Recurrent Neural Networks: The 30M Factoid Question-Answer Corpus
{I. V. Serban, A. G. Duran}, C. Gulcehre, S. Ahn, S. Chandar, A. Courville, Y. Bengio
ACL 16

Scalable MCMC for Mixed Membership Stochastic Blockmodels
{W. Li, S. Ahn}, and M. Welling
AISTATS 16

Scalable Overlapping Community Detection
I. El-Helw, R. Hofman, W. Li, S. Ahn, M. Welling, H. Bal
ParLearning 16, Best Paper Award

Learning Latent Multiscale Structure using Recurrent Neural Networks
J. Chung, S. Ahn, Y. Bengio
NIPS 16 Workshop on Neural Abstract Machines & Program Induction (NAMPI)

A Neural Knowledge Language Model
S. Ahn, H. Choi, T. Parnamaa, Y. Bengio
arXiv

Hierarchical Memory Networks
S. Chandar, S. Ahn, H. Larochelle, P. Vincent, G. Tasauro, Y. Bengio
arXiv

~2015 (selected publications)

Stochastic Gradient MCMC: Algorithms and Applications
PhD Dissertation

Large-Scale Distributed Bayesian Matrix Factorization using Stochastic Gradient MCMC
S. Ahn, A. Korattikara, N. Liu, S. Rajan, and M. Welling
KDD 15

Distributed Stochastic Gradient MCMC
S. Ahn, B. Shahbaba, and M. Welling
ICML 14

Distributed and Adaptive Darting Monte Carlo through Regenerations
S. Ahn, Y. Chen, and M. Welling
AISTATS 13

Bayesian Posterior Sampling via Stochastic Gradient Fisher Scoring
S. Ahn, A. Korattikara, and M. Welling
ICML 12 Best Paper Award