Publications | Minseok Jeong

2026

Mitigating Overgeneralization in RND via Spectral Target Design

Minseok Jeong^*, Yechan Lee^*, Hyewon Choi, and 2 more authors

In Proceedings of the Conference on Neural Information Processing Systems (NeurIPS), 2026

Abs

Random Network Distillation (RND) is a scalable novelty signal for RL based on a fixed random target network and a trained predictor. However, it can overgeneralize: the predictor extrapolates the random target off the data manifold, causing novelty scores to collapse on unfamiliar inputs. This paper analyzes this failure through a kernel-theoretic lens. Under a GP target model and an NTK-regime predictor, we show that the RND novelty score is a cross-kernel residual variance: it is governed jointly by the target covariance kernel and the predictor NTK, rather than by either one alone. This identifies overgeneralization as spectral under-excitation. Motivated by this view, we formulate spectral target design as a principle for shaping RND novelty geometry. As a concrete instance, we replace the implicit random-network target with a bandwidth-controlled Random Fourier Feature (RFF) target, redistributing residual signal toward modes less attenuated by the predictor. Experiments on offline D4RL, online Atari, and Fetch manipulation benchmarks show improved novelty discrimination and downstream RL performance.
Scalable Infinitesimal Generator–Based Koopman Learning for Long-Horizon Prediction

Minseok Jeong, SooJean Han, and Hyosang Shin

In Proceedings of the Learning for Dynamics and Control Conference (L4DC), 2026

Abs

Koopman operator theory provides a linear representation of nonlinear dynamics, offering strong potential for long-horizon prediction. However, most existing methods rely on snapshot-based one-step fitting and often suffer from error accumulation over extended rollouts. To address this issue, prior efforts have explored extending the training horizon or developing model-based approaches that leverage physics-informed formulations—such as infinitesimal generator fitting—to better exploit the underlying system dynamics. In this paper, we show that local transition fitting itself fundamentally limits long-horizon stability, whereas generator-based learning provides a more principled alternative. Despite their reliable advantages, generator-based learning schemes remain computationally demanding, as training dense MLP-based dictionaries with Jacobian computations is often intractable in practice. Building upon this line of work, we introduce a Random Fourier Feature–lifted Koopman formulation (RFF-PIKN) that employs shift-invariant RKHS representations for scalable and stable approximation, reducing computational cost under repeated resampling. We theoretically show that RFF-PIKN converges to a local minimizer of the true population generator risk, enabling accurate and stable long-horizon prediction. These results highlight the promise of RKHS-based physics-informed Koopman learning in advancing model-based prediction frameworks.

2025

Mitigating Instability in High Residual Adaptive Sampling for PINNs via Langevin Dynamics

Minseok Jeong^*, Giup Seo^*, and Euiseok Hwang

In Proceedings of the Conference on Neural Information Processing Systems (NeurIPS), 2025

Spotlight

Abs

Recently, physics-informed neural networks (PINNs) have gained attention in the scientific community for their potential to solve partial differential equations (PDEs). However, they face challenges related to resource efficiency and slow convergence. Adaptive sampling methods, which prioritize collocation points with high residuals, improve both efficiency and accuracy. However, these methods often neglect points with medium or low residuals, which can affect stability as the complexity of the model increases. In this paper, we investigate this limitation and show that high residual-based approaches require stricter learning rate bounds to ensure stability. To address this, we propose a Langevin dynamics-based Adaptive Sampling (LAS) framework that is robust to various learning rates and model complexities. Our experiments demonstrate that the proposed method outperforms existing approaches in terms of relative error, and stability across a range of enviroments, including high-dimensional PDEs where Monte Carlo integration-based methods typically suffer from instability.

2024

Geomixup: Robust Radio Frequency Based Indoor Localization Leveraging Geographic Information

Minseok Jeong, Giup Seo, and Euiseok Hwang

In 2024 IEEE International Conference on Communications Workshops (ICC Workshops), 2024

Abs DOI

In this paper, we address the problem of non-robustness in empirical risk minimization-based learning schemes for indoor localization using radio frequency (RF) fingerprints. To overcome this challenge, we propose a novel solution that leverages the manifold structure of the channel state information (CSI) data, exploiting its geographical similarity. Our approach involves a simple yet effective data augmentation scheme named \textitGeomixup, which utilizes geographic information to synthesize data for training. By incorporating geolocation-based data augmentation, we enable robust localization based on CSI, ensuring that the model’s decisions are grounded in the highly correlated geolocation feature, rather than weakly correlated ones. To evaluate the performance of the proposed method, we utilized two performance measures, \textitpseudo accuracy and \textitdistance error, on the data points that were not encountered during training. They consider the confidence assigned to each location prediction and apply a larger penalty in cases where the model exhibits high confidence in misplacement. Through extensive RF-based localization tests, we demonstrate that our proposed scheme outperforms other learning methods in terms of robustness, all while maintaining a negligible loss of accuracy. This significant improvement in robustness ensures the reliability and stability of our approach, making it highly suitable for real-world indoor localization applications.

2022

Age of Information Optimization by Deep Reinforcement Learning for Random Access in Machine Type Communication

Minseok Jeong, Giup Seo, and Euiseok Hwang

In 2022 IEEE International Conference on Big Data (Big Data), 2022

Abs DOI

For machine type communication with random access (RA) protocol, finding optimal policy using deep reinforcement learning (DRL) is being actively investigated for various quality of service requirements. In particular, it was shown that throughput-based reward function in DRL can derive policies that outperform conventional exponential backoff (EB)-based algorithms in terms of throughput and fairness. However, age of information (AoI), which is a measure of the freshness of data, was not addressed in the process of training the DRL agent and only used as a measure of fairness. It has been theoretically proven that, even in the simplest queuing system, maximizing throughput does not guarantee minimizing AoI. In this paper, we proposed a novel DRL scheme to directly optimize AoI in a slotted ALOHA RA channel. By taking into account urgency and packet age as extra local information for a reward, AoI could be improved while preserving the throughput. Numerical simulations showed that the proposed DRL approach could achieve an improvement of 22.04% in AoI performance, with a marginal throughput loss of around 3.95%, compared to the existing throughput-based DRL method.