Jea (Jay) Kwon

Max Planck Institute for Security and Privacy

Universitaetsstr. 140

44799 Bochum, Germany

I am a postdoctoral researcher at the Max Planck Institute for Security and Privacy (MPI-SP), working on machine memory, mechanistic interpretability, and alignment in large language models.

My path to AI began in neuroscience wet labs—recording synaptic responses, inducing LTP and LTD to study memory formation, and using optogenetics to causally link neural circuits to behavior. I then moved upstream, developing AI systems to analyze the complex behaviors these manipulations produced. This arc—from probing biological memory, to controlling it, to quantifying its behavioral consequences—now shapes how I approach neural networks.

I study AI Engrams: locating where learned knowledge resides within model parameters and building precise methods to edit or erase it. I am also interested in the correspondence between biological and artificial neural networks—reinterpreting transformers as models of hippocampal memory consolidation and studying representational alignment between visual cortex and convolutional networks.

My underlying conviction is that interpretability, steerability, and alignment form a causal chain. If we can precisely locate memory traces within parameters, we gain the ability to steer model behavior at its source. And if we can steer it, we can align it. The path to trustworthy AI runs through understanding the learning and memory of AI.

Ph.D. from Korea University; previously at IBS and KIST.

news

Apr 06, 2026	Two papers accepted at ACL 2026: “Exploring LLM Behavior in Relational Moral Dilemmas: Moral Rightness, Predicted Human Behavior, and Model Decisions” (Findings) and “How Training Data Shapes the Use of Parametric and In-Context Knowledge in Language Models” (Main).
Apr 01, 2026	Honored to announce that I am organizing a KSBNS symposium, “Minds Meet Machines: Bidirectional Innovations in AI and Neuroscience”.
Feb 18, 2026	Our paper “Cerebellar tonic inhibition orchestrates the maturation of information processing and motor coordination” has been accepted at Experimental & Molecular Medicine.
Jan 22, 2026	Two papers accepted at ICLR 2026: “Bilinear Relational Structure Fixes Reversal Curse and Enables Consistent Model Editing” and “Erase or Hide? Suppressing Spurious Unlearning Neurons for Robust Unlearning.”
Nov 17, 2025	Our paper “Dropouts in Confidence: Moral Uncertainty in Human-LLM Alignment” has been accepted at AAAI 2026.
Aug 21, 2025	Glad to contribute in this paper publshied at Nature Communications, “Integrating Artificial Intelligence and Optogenetics for Parkinson’s Disease Diagnosis”.
Apr 01, 2025	Honored to announce that I am co-organizing a KSBNS & CJK symposium, “Neuroscience-inspired AI: Computational insights into biological and artificial intelligence”.
Jan 22, 2025	Our paper “Brain-inspired Lp-Convolution Benefits Large Kernels and Aligns Better with Visual Cortex” has been accepted at ICLR 2025.
Nov 01, 2024	Joined Max Planck Institute for Security and Privacy (MPI-SP) as a Postdoctoral Researcher.
May 20, 2024	Our paper “SUBTLE: An Unsupervised Platform with Temporal Link Embedding that Maps Animal Behavior” has been accepted at International Journal of Computer Vision.
Nov 24, 2023	Honored to receive the JKAIA 2023 Excellence Paper Award for our work, “Brain-inspired Lp-convolution benefits large kernels and aligns better with visual cortex”.
Sep 21, 2023	Our paper “Transformer as a Hippocampal Memory Consolidation Model Based on NMDAR-Inspired Nonlinearity” has been accepted at NeurIPS 2023.
May 21, 2022	Honored to receive the KSBNS 2022 Best Presentation Award for “ABCD-analysis: Mapping animal behavior and differential analysis of kinematic features without selection bias”.

selected publications

EMM

Cerebellar Tonic Inhibition Orchestrates the Maturation of Information Processing and Motor Coordination

Jea Kwon^*, Sunpil Kim^*, Junsung Woo, and 5 more authors

Experimental & Molecular Medicine, 2026

Bib

@article{kwon2026emm,
  title = {Cerebellar Tonic Inhibition Orchestrates the Maturation of Information Processing and Motor Coordination},
  author = {Kwon, Jea and Kim, Sunpil and Woo, Junsung and Tanaka-Yamamoto, Keiko and James, Olivia and De Schutter, Erik and Hong, Sungho and Lee, C. Justin},
  journal = {Experimental \& Molecular Medicine},
  year = {2026},
  url = {https://doi.org/10.1038/s12276-026-01657-8}
}

AAAI

Dropouts in Confidence: Moral Uncertainty in Human-LLM Alignment

Jea Kwon^*, Luiz Felipe Vecchietti, Sungwon Park, and 1 more author

In AAAI Conference on Artificial Intelligence, 2026

Bib

@inproceedings{kwon2026aaai,
  title = {Dropouts in Confidence: Moral Uncertainty in Human-LLM Alignment},
  author = {Kwon, Jea and Vecchietti, Luiz Felipe and Park, Sungwon and Cha, Meeyoung},
  booktitle = {AAAI Conference on Artificial Intelligence},
  year = {2026},
}

ICLR

Brain-inspired Lp-Convolution Benefits Large Kernels and Aligns Better with Visual Cortex

Jea Kwon^*, SungJun Lim, Kyungwoo Song, and 1 more author

In International Conference on Learning Representations, 2025

Bib Code

@inproceedings{kwon2025iclr,
  title = {Brain-inspired Lp-Convolution Benefits Large Kernels and Aligns Better with Visual Cortex},
  author = {Kwon, Jea and Lim, SungJun and Song, Kyungwoo and Lee, C. Justin},
  booktitle = {International Conference on Learning Representations},
  year = {2025},
  url = {https://openreview.net/forum?id=0LSAmFCc4p}
}

IJCV

SUBTLE: An Unsupervised Platform with Temporal Link Embedding that Maps Animal Behavior

Jea Kwon^*, Sunpil Kim, Dong-Kyum Kim, and 4 more authors

International Journal of Computer Vision, 2024

Bib Code

@article{kwon2024ijcv,
  title = {SUBTLE: An Unsupervised Platform with Temporal Link Embedding that Maps Animal Behavior},
  author = {Kwon, Jea and Kim, Sunpil and Kim, Dong-Kyum and Joo, Jinhyeong and Kim, SoHyung and Cha, Meeyoung and Lee, C. Justin},
  journal = {International Journal of Computer Vision},
  year = {2024},
  url = {https://doi.org/10.1007/s11263-024-02072-0}
}

Exp. Neurobiol.

Egocentric 3D Skeleton Learning in a Deep Neural Network Encodes Obese-like Motion Representations

Jea Kwon^*, Moonsun Sa, Hyewon Kim, and 2 more authors

Experimental Neurobiology, 2024

Bib

@article{kwon2024expneurobiol,
  title = {Egocentric 3D Skeleton Learning in a Deep Neural Network Encodes Obese-like Motion Representations},
  author = {Kwon, Jea and Sa, Moonsun and Kim, Hyewon and Seong, Yejin and Lee, C. Justin},
  journal = {Experimental Neurobiology},
  year = {2024},
  url = {https://doi.org/10.5607/en24008}
}

NeurIPS

Transformer as a Hippocampal Memory Consolidation Model Based on NMDAR-Inspired Nonlinearity

Dong-Kyum Kim^*, Jea Kwon^*, Meeyoung Cha, and 1 more author

In Advances in Neural Information Processing Systems, 2023

Bib

@inproceedings{kim2023neurips,
  title = {Transformer as a Hippocampal Memory Consolidation Model Based on NMDAR-Inspired Nonlinearity},
  author = {Kim, Dong-Kyum and Kwon, Jea and Cha, Meeyoung and Lee, C. Justin},
  booktitle = {Advances in Neural Information Processing Systems},
  year = {2023},
  url = {https://openreview.net/forum?id=vKpVJxplmB}
}

Mol. Brain

Retina-Attached Slice Recording Reveals Light-Triggered Tonic GABA Signaling in Suprachiasmatic Nucleus

Jea Kwon^*, Minwoo Wendy Jang, and C. Justin Lee

Molecular Brain, 2021

Bib

@article{kwon2021molbrain,
  title = {Retina-Attached Slice Recording Reveals Light-Triggered Tonic GABA Signaling in Suprachiasmatic Nucleus},
  author = {Kwon, Jea and Jang, Minwoo Wendy and Lee, C. Justin},
  journal = {Molecular Brain},
  year = {2021},
  url = {https://doi.org/10.1186/s13041-021-00881-9}
}