arXiv subjects

Jungmin Lee

Publications and source records attributed to Jungmin Lee.

13 recordsLinked to original sources

arXiv2026

12 nm FinFETs from 292 K to 10 mK: Characterization and a Temperature-Continuous Compact Model for Qubit Control

We report cryogenic characterization and compact modeling of GlobalFoundries (GF) 12LP regular-voltage-threshold (RVT) n- and p-FinFETs, measured from 292 K to 10 mK. Both devices retain sub-Kelvin gate control without carrier freeze-out, with the n-type device exhibiting an intrinsic sub-threshold slope (SS) floor of 24.8 mV/dec. To bypass high on-chip series routing resistance that confounds peak-gm extraction, we adopt a fixed-current (3.26 uA/um) methodology. These low-current metrics inform a continuous 1.36 K to 292 K compact model using dynamic parameter injection, preserving room-temperature process development kit (PDK) qualification.

physics.gen-ph↗

arXiv2026

Cognitive Ability and Tournament Entry: Evidence from Three Korean Populations

We compare tournament entry among three Korean groups raised in different institutional environments: South Korea, North Korea, and China. Experiments with a random bonus show that North Korean (NK) refugees enter tournaments less often than South Korean (SK) participants and Korean-Chinese immigrants. The conditional NK-SK difference becomes statistically insignificant when Raven score is included. A choice model with probability weighting suggests that lower cognitive ability is associated with lower expected performance, more pessimistic beliefs, and, within the model, greater aversion to competition.

econ.GN↗

arXiv2026

Rethinking Detection Calibration: A Coordinate and Direction Perspective

Deep learning based object detectors require trustworthiness beyond competitive detection performance, but deep neural networks are prone to overconfident predictions, assigning high confidence scores to predictions that are likely to be inaccurate. To improve the alignment between confidence scores and prediction accuracy, existing methods calibrate confidence scores based on box-level localization, such as precision or intersection over union with the ground truth bounding box. However, box-level localization reflects only a measure of agreement between the predicted box and the ground truth, resulting in calibrated confidence scores for box-level accuracy failing to capture the localization accuracy of coordinates of box. To tackle this issue, we propose a novel post-hoc calibration framework, rethinking detection calibration (ReDC), which provides reliable coordinate-level confidence scores, including directional information. The proposed framework defines coordinate-wise alignment and deviation direction between predictions and ground truth. Based on the alignment measure, confidence re-encoding produces reliable coordinate-level confidence scores, while directional displacement estimation predicts coordinate-wise deviation directions. Extensive experiments under in-domain and out-domain scenarios demonstrate that the proposed approach expresses the coordinate-wise localization of detected objects more precisely than existing methods. Furthermore, our method covers the representational scope of prior calibration approaches by aggregating coordinate-level confidence scores into box-level localization.

cs.CV↗

arXiv2026

Inverse Design of Multi-Layered Manufacturable Pixelated Diplexers Through Optimized Geometrical Configuration and Meshing Strategy in MoM

This paper presents a fast inverse design framework for complex multilayered, multiport pixelated surfaces - a class of structures largely unexplored in current research. Leveraging a method-of-moments (MoM) electromagnetic (EM) solver, the framework enables the rapid synthesis of pixelated device designs. A novel matrix reconstruction technique, based on pre-labeling matrix entries as "inter-pixel" or "inner-pixel," accelerates simulations for each variation of the pixelated structure. To mitigate the cubic increase in computation time associated with additional layers, GPU acceleration is employed. Further enhancing convergence speed, a stochastic multi-pixel flipping search algorithm is integrated into the framework. The effectiveness of this approach is demonstrated through the design of a diplexer achieving a -3-dB bandwidth for one channel spanning 5.23-5.94 GHz and another covering 6.17-7.15 GHz, validated by a full-wave solver.

physics.app-ph↗

arXiv2026

LeagueBot: A Voice LLM Companion of Cognitive and Emotional Support for Novice Players in Competitive Games

Competitive games pose steep learning curves and strong social pressures, often discouraging novice players and limiting sustained engagement. To address these challenges, this study introduces LeagueBot, a large language model-based voice chatbot designed to provide both informational and emotional support during live gameplay in league of legends, one of the most competitive multiplayer online battle arena games. In a within-subjects experiment with 33 novice players, LeagueBot was found to reduce cognitive challenge, performative challenge, and perceived tension. Qualitative analysis further identified three themes: enhanced access to game information, relief from cognitive burden, and practical limitations. Participants noted that LeagueBot offered context-appropriate guidance and emotional support, helping ease the steep learning curve and psychological pressures of competitive gaming. Together, these findings underscore the potential of voice-based LLM companions to assist novice players in competitive environments and highlight their broader applicability for real-time support in other high-pressure contexts.

cs.HC↗

arXiv2025

The Effect of Empathic Expression Levels in Virtual Human Interaction: A Controlled Experiment

As artificial intelligence (AI) systems become increasingly embedded in everyday life, the ability of interactive agents to express empathy has become critical for effective human-AI interaction, particularly in emotionally sensitive contexts. Rather than treating empathy as a binary capability, this study examines how different levels of empathic expression in virtual human interaction influence user experience. We conducted a between-subject experiment (n = 70) in a counseling-style interaction context, comparing three virtual human conditions: a neutral dialogue-based agent, a dialogue-based empathic agent, and a video-based empathic agent that incorporates users' facial cues. Participants engaged in a 15-minute interaction and subsequently evaluated their experience using subjective measures of empathy and interaction quality. Results from analysis of variance (ANOVA) revealed significant differences across conditions in affective empathy, perceived naturalness of facial movement, and appropriateness of facial expression. The video-based empathic expression condition elicited significantly higher affective empathy than the neutral baseline (p < .001) and marginally higher levels than the dialogue-based condition (p < .10). In contrast, cognitive empathy did not differ significantly across conditions. These findings indicate that empathic expression in virtual humans should be conceptualized as a graded design variable, rather than a binary capability, with visually grounded cues playing a decisive role in shaping affective user experience.

cs.HC↗

arXiv2025

Leveraging KV Similarity for Online Structured Pruning in LLMs

Pruning has emerged as a promising direction for accelerating large language model (LLM) inference, yet existing approaches often suffer from instability because they rely on offline calibration data that may not generalize across inputs. In this work, we introduce Token Filtering, a lightweight online structured pruning technique that makes pruning decisions directly during inference without any calibration data. The key idea is to measure token redundancy via joint key-value similarity and skip redundant attention computations, thereby reducing inference cost while preserving critical information. To further enhance stability, we design a variance-aware fusion strategy that adaptively weights key and value similarity across heads, ensuring that informative tokens are retained even under high pruning ratios. This design introduces no additional memory overhead and provides a more reliable criterion for token importance. Extensive experiments on LLaMA-2 (7B/13B), LLaMA-3 (8B), and Mistral (7B) demonstrate that Token Filtering consistently outperforms prior structured pruning methods, preserving accuracy on commonsense reasoning benchmarks and maintaining strong performance on challenging tasks such as MMLU, even with 50% pruning.

cs.CL↗

arXiv2025

InsideOut: Integrated RGB-Radiative Gaussian Splatting for Comprehensive 3D Object Representation

We introduce InsideOut, an extension of 3D Gaussian splatting (3DGS) that bridges the gap between high-fidelity RGB surface details and subsurface X-ray structures. The fusion of RGB and X-ray imaging is invaluable in fields such as medical diagnostics, cultural heritage restoration, and manufacturing. We collect new paired RGB and X-ray data, perform hierarchical fitting to align RGB and X-ray radiative Gaussian splats, and propose an X-ray reference loss to ensure consistent internal structures. InsideOut effectively addresses the challenges posed by disparate data representations between the two modalities and limited paired datasets. This approach significantly extends the applicability of 3DGS, enhancing visualization, simulation, and non-destructive testing capabilities across various domains.

cs.CV↗

arXiv2025

Persode: Personalized Visual Journaling with Episodic Memory-Aware AI Agent

Reflective journaling often lacks personalization and fails to engage Generation Alpha and Z, who prefer visually immersive and fast-paced interactions over traditional text-heavy methods. Visual storytelling enhances emotional recall and offers an engaging way to process personal expe- riences. Designed with these digital-native generations in mind, this paper introduces Persode, a journaling system that integrates personalized onboarding, memory-aware conversational agents, and automated visual storytelling. Persode captures user demographics and stylistic preferences through a tailored onboarding process, ensuring outputs resonate with individual identities. Using a Retrieval-Augmented Generation (RAG) framework, it prioritizes emotionally significant memories to provide meaningful, context-rich interactions. Additionally, Persode dynamically transforms user experiences into visually engaging narratives by generating prompts for advanced text-to-image models, adapting characters, backgrounds, and styles to user preferences. By addressing the need for personalization, visual engagement, and responsiveness, Persode bridges the gap between traditional journaling and the evolving preferences of Gen Alpha and Z.

cs.HC↗

arXiv2025

Group-wise Scaling and Orthogonal Decomposition for Domain-Invariant Feature Extraction in Face Anti-Spoofing

Domain Generalizable Face Anti-Spoofing (DGFAS) methods effectively capture domain-invariant features by aligning the directions (weights) of local decision boundaries across domains. However, the bias terms associated with these boundaries remain misaligned, leading to inconsistent classification thresholds and degraded performance on unseen target domains. To address this issue, we propose a novel DGFAS framework that jointly aligns weights and biases through Feature Orthogonal Decomposition (FOD) and Group-wise Scaling Risk Minimization (GS-RM). Specifically, GS-RM facilitates bias alignment by balancing group-wise losses across multiple domains. FOD employs the Gram-Schmidt orthogonalization process to decompose the feature space explicitly into domain-invariant and domain-specific subspaces. By enforcing orthogonality between domain-specific and domain-invariant features during training using domain labels, FOD ensures effective weight alignment across domains without negatively impacting bias alignment. Additionally, we introduce Expected Calibration Error (ECE) as a novel evaluation metric for quantitatively assessing the effectiveness of our method in aligning bias terms across domains. Extensive experiments on benchmark datasets demonstrate that our approach achieves state-of-the-art performance, consistently improving accuracy, reducing bias misalignment, and enhancing generalization stability on unseen target domains.

cs.CV↗

arXiv2023

Implicit Bias against a Capitalistic Society Predicts Market Earnings

This paper investigates whether ideological indoctrination by living in a communist regime relates to low economic performance in a market economy. We recruit North Korean refugees and measure their implicit bias against South Korea by using the Implicit Association Test. Conducting double auction and bilateral bargaining market experiments, we find that North Korean refugees with a larger bias against the capitalistic society have lower expectations about their earning potential, exhibit trading behavior with lower target profits, and earn less profits. These associations are robust to conditioning on correlates of preferences, human capital, and assimilation experiences.

econ.GN↗

arXiv2022

Lightweight Image Enhancement Network for Mobile Devices Using Self-Feature Extraction and Dense Modulation

Convolutional neural network (CNN) based image enhancement methods such as super-resolution and detail enhancement have achieved remarkable performances. However, amounts of operations including convolution and parameters within the networks cost high computing power and need huge memory resource, which limits the applications with on-device requirements. Lightweight image enhancement network should restore details, texture, and structural information from low-resolution input images while keeping their fidelity. To address these issues, a lightweight image enhancement network is proposed. The proposed network include self-feature extraction module which produces modulation parameters from low-quality image itself, and provides them to modulate the features in the network. Also, dense modulation block is proposed for unit block of the proposed network, which uses dense connections of concatenated features applied in modulation layers. Experimental results demonstrate better performance over existing approaches in terms of both quantitative and qualitative evaluations.

eess.IV↗

arXiv2018

Attention-based Ensemble for Deep Metric Learning

Deep metric learning aims to learn an embedding function, modeled as deep neural network. This embedding function usually puts semantically similar images close while dissimilar images far from each other in the learned embedding space. Recently, ensemble has been applied to deep metric learning to yield state-of-the-art results. As one important aspect of ensemble, the learners should be diverse in their feature embeddings. To this end, we propose an attention-based ensemble, which uses multiple attention masks, so that each learner can attend to different parts of the object. We also propose a divergence loss, which encourages diversity among the learners. The proposed method is applied to the standard benchmarks of deep metric learning and experimental results show that it outperforms the state-of-the-art methods by a significant margin on image retrieval tasks.

cs.CV↗