arXiv subjects

Jiwan Kim

Publications and source records attributed to Jiwan Kim.

17 recordsLinked to original sources

arXiv2026

Three-temperature atomistic spin-lattice dynamics in LAMMPS: a moment-consistent, fluctuation-dissipation-correct extension and its validation on ultrafast demagnetization and all-optical switching of GdFeCo

Atomistic spin dynamics (ASD) codes like VAMPIRE simulate femtosecond-laser-induced ultrafast demagnetization and all-optical switching (AOS) on a rigid lattice; the SPIN package of the Large-scale Atomic/Molecular Massively Parallel Simulator (LAMMPS) instead propagates spins and lattice together, but until now could couple its spin thermostat only to a single global temperature, precluding three-temperature (3TM) simulations. We present a validated 3TM extension with two new fixes: langevin/spin/ttm, coupling the stochastic spin bath to the local electron-temperature field of fix ttm; and moment/scale/spin, supplying the per-atom moment-dependent ($1/\mu_i$) prefactor required for heterogeneous-moment systems, with deterministic terms scaling as $1/\mu_i$ and, as required by the fluctuation-dissipation theorem, stochastic noise scaling as $1/\sqrt{\mu_i}$. We validate the framework on single-species benchmarks-precession, the bcc-Fe Curie curve, a continuous demagnetization-remagnetization-precession trajectory, and a lattice-strain acoustic-phonon pulse unavailable to spin-only ASD codes-then on heat-induced AOS of GdFeCo with literature parameters (Radu/Ostler exchange constants, moments, and damping). The corrected integrator reproduces the experimentally reported transient ferromagnetic-like sublattice alignment, field-insensitive thermal switching, a switching probability approaching unity across a genuine 8x8 literature damping phase diagram, a non-monotonic critical-cooling-duration boundary that disappears above $T_e^0 \approx 2000 \text{ K}$, a composition-dependent switching window centered on angular-momentum compensation, and convergence in system size with monotonic, physical damping dependence. These results establish LAMMPS as a quantitatively validated platform for three-temperature spin-lattice simulations of ultrafast magnetism.

cond-mat.mtrl-sci↗

arXiv2026

Sink-Token-Aware Pruning for Fine-Grained Video Understanding in Efficient Video LLMs

Video Large Language Models (Video LLMs) incur high inference latency due to a large number of visual tokens provided to LLMs. To address this, training-free visual token pruning has emerged as a solution to reduce computational costs; however, existing methods are primarily validated on Multiple-Choice Question Answering (MCQA) benchmarks, where coarse-grained cues often suffice. In this work, we reveal that these methods suffer a sharp performance collapse on fine-grained understanding tasks requiring precise visual grounding, such as hallucination evaluation. To explore this gap, we conduct a systematic analysis and identify sink tokens--semantically uninformative tokens that attract excessive attention--as a key obstacle to fine-grained video understanding. When these sink tokens survive pruning, they distort the model's visual evidence and hinder fine-grained understanding. Motivated by these insights, we propose Sink-Token-aware Pruning (SToP), a simple yet effective plug-and-play method that introduces a sink score to quantify each token's tendency to behave as a sink and applies this score to existing spatial and temporal pruning methods to suppress them, thereby enhancing video understanding. To validate the effectiveness of SToP, we apply it to state-of-the-art pruning methods (VisionZip, FastVid, and Holitom) and evaluate it across diverse benchmarks covering hallucination, open-ended generation, compositional reasoning, and MCQA. Our results demonstrate that SToP significantly boosts performance, even when pruning up to 90% of visual tokens.

cs.LG↗

arXiv2026

Why and When Visual Token Pruning Fails? A Study on Relevant Visual Information Shift in MLLMs Decoding

Recently, visual token pruning has been studied to handle the vast number of visual tokens in Multimodal Large Language Models. However, we observe that while existing pruning methods perform reliably on simple visual understanding, they struggle to effectively generalize to complex visual reasoning tasks, a critical gap underexplored in previous studies. Through a systematic analysis, we identify Relevant Visual Information Shift (RVIS) during decoding as the primary failure driver. To address this, we propose Decoding-stage Shift-aware Token Pruning (DSTP), a training-free add-on framework that enables existing pruning methods to align visual tokens with shifting reasoning requirements during the decoding stage. Extensive experiments demonstrate that DSTP significantly mitigates performance degradation of pruning methods in complex reasoning tasks, while consistently yielding performance gains even across visual understanding benchmarks. Furthermore, DSTP demonstrates effectiveness across diverse state-of-the-art architectures, highlighting its generalizability and efficiency with minimal computational overhead.

cs.CV↗

arXiv2026

WatchHand: Enabling Continuous Hand Pose Tracking On Off-the-Shelf Smartwatches

Tracking hand poses on wrist-wearables enables rich, expressive interactions, yet remains unavailable on commercial smartwatches, as prior implementations rely on external sensors or custom hardware, limiting their real-world applicability. To address this, we present WatchHand, the first continuous 3D hand pose tracking system implemented on off-the-shelf smartwatches using only their built-in speaker and microphone. WatchHand emits inaudible frequency-modulated continuous waves and captures their reflections from the hand. These acoustic signals are processed by a deep-learning model that estimates 3D hand poses for 20 finger joints. We evaluate WatchHand across diverse real-world conditions -- multiple smartwatch models, wearing-hands, body postures, noise conditions, pose-variation protocols -- and achieve a mean per-joint position error of 7.87 mm in cross-session tests with device remounting. Although performance drops for unseen users or gestures, the model adapts effectively with lightweight fine-tuning on small amounts of data. Overall, WatchHand lowers the barrier to smartwatch-based hand tracking by eliminating additional hardware while enabling robust, always-available interactions on millions of existing devices.

cs.HC↗

arXiv2025

Adaptive Graph Rewiring to Mitigate Over-Squashing in Mesh-Based GNNs for Fluid Dynamics Simulations

Mesh-based simulation using Graph Neural Networks (GNNs) has been recognized as a promising approach for modeling fluid dynamics. However, the mesh refinement techniques which allocate finer resolution to regions with steep gradients can induce the over-squashing problem in mesh-based GNNs, which prevents the capture of long-range physical interactions. Conventional graph rewiring methods attempt to alleviate this issue by adding new edges, but they typically complete all rewiring operations before applying them to the GNN. These approaches are physically unrealistic, as they assume instantaneous interactions between distant nodes and disregard the distance information between particles. To address these limitations, we propose a novel framework, called Adaptive Graph Rewiring in Mesh-Based Graph Neural Networks (AdaMeshNet), that introduces an adaptive rewiring process into the message-passing procedure to model the gradual propagation of physical interactions. Our method computes a rewiring delay score for bottleneck nodes in the mesh graph, based on the shortest-path distance and the velocity difference. Using this score, it dynamically selects the message-passing layer at which new edges are rewired, which can lead to adaptive rewiring in a mesh graph. Extensive experiments on mesh-based fluid simulations demonstrate that AdaMeshNet outperforms conventional rewiring methods, effectively modeling the sequential nature of physical interactions and enabling more accurate predictions.

cs.LG↗

arXiv2025

CompoDistill: Attention Distillation for Compositional Reasoning in Multimodal LLMs

Recently, efficient Multimodal Large Language Models (MLLMs) have gained significant attention as a solution to their high computational complexity, making them more practical for real-world applications. In this regard, the knowledge distillation (KD) approach has emerged as a promising alternative, which transfers the rich visual and linguistic knowledge from a larger model (teacher) to a smaller model (student). However, we observe that existing KD methods struggle to effectively distill the teacher MLLM's rich visual perception abilities to the student, a challenge that has been largely overlooked in previous studies. Through a systematic analysis, we identify visual attention misalignment between student and teacher as the main cause of this issue. Based on this insight, we propose CompoDistill, a novel KD framework that explicitly aligns the student's visual attention with that of the teacher to enhance the student's visual perception abilities. Our extensive experiments show that CompoDistill significantly improves performance on compositional reasoning tasks that require visual perception abilities while maintaining strong performance on visual question answering tasks, as done in existing studies. Furthermore, CompoDistill demonstrates effectiveness with a more advanced backbone, highlighting its generalizability.

cs.CV↗

arXiv2025

Planar Ballistic Electron Emission Spectroscopy for Single-Shot Probing of Energy Barrier Inhomogeneity at Junction Interface

We propose an experimental methodology for probing the energy barrier inhomogeneity at the metal/semiconductor interface without the need for time-consuming microscopic survey. It is based on the known statistical nature of the interfacial energy barrier and the use of planar tunnel junction as an array of parallelly-connected ballistic electron emission microscopy (BEEM) tips. In order to analyze a lump of local BEEM signals, we incorporate the Tung model into the Bell-Kaiser theory. To validate our theoretical strategies, we investigate the interfacial energy barrier inhomogeneity of Pt/4H-SiC(0001) junction as a model system.

cond-mat.mes-hall↗

arXiv2025

EchoForce: Continuous Grip Force Estimation from Skin Deformation Using Active Acoustic Sensing on a Wristband

Grip force is commonly used as an overall health indicator in older adults and is valuable for tracking progress in physical training and rehabilitation. Existing methods for wearable grip force measurement are cumbersome and user-dependent, making them insufficient for practical, continuous grip force measurement. We introduce EchoForce, a novel wristband using acoustic sensing for low-cost, non-contact measurement of grip force. EchoForce captures acoustic signals reflected from subtle skin deformations by flexor muscles on the forearm. In a user study with 11 participants, EchoForce achieved a fine-tuned user-dependent mean error rate of 9.08% and a user-independent mean error rate of 12.3% using a foundation model. Our system remained accurate between sessions, hand orientations, and users, overcoming a significant limitation of past force sensing systems. EchoForce makes continuous grip force measurement practical, providing an effective tool for health monitoring and novel interaction techniques.

cs.HC↗

arXiv2025

Garibaldi: A Pairwise Instruction-Data Management for Enhancing Shared Last-Level Cache Performance in Server Workloads

Modern CPUs suffer from the frontend bottleneck because the instruction footprint of server workloads exceeds the private cache capacity. Prior works have examined the CPU components or private cache to improve the instruction hit rate. The large footprint leads to significant cache misses not only in the core and faster-level cache but also in the last-level cache (LLC). We observe that even with an advanced branch predictor and instruction prefetching techniques, a considerable amount of instruction accesses descend to the LLC. However, state-of-the-art LLC designs with elaborate data management overlook handling the instruction misses that precede corresponding data accesses. Specifically, when an instruction requiring numerous data accesses is missed, the frontend of a CPU should wait for the instruction fetch, regardless of how much data are present in the LLC. To preserve hot instructions in the LLC, we propose Garibaldi, a novel pairwise instruction-data management scheme. Garibaldi tracks the hotness of instruction accesses by coupling it with that of data accesses and adopts management techniques. On the one hand, this scheme includes a selective protection mechanism that prevents the cache evictions of high-cost instruction cachelines. On the other hand, in the case of unprotected instruction line misses, Garibaldi conservatively issues prefetch requests of the paired data lines while handling those misses. In our experiments, we evaluate Garibaldi with 16 server workloads on a 40-core machine. We also implement Garibaldi on top of a modern LLC design, including Mockingjay. Garibaldi improves 13.2% and 6.1% of CPU performance on baseline LLC design and Mockingjay, respectively.

cs.AR↗

arXiv2025

Disentangling and Generating Modalities for Recommendation in Missing Modality Scenarios

Multi-modal recommender systems (MRSs) have achieved notable success in improving personalization by leveraging diverse modalities such as images, text, and audio. However, two key challenges remain insufficiently addressed: (1) Insufficient consideration of missing modality scenarios and (2) the overlooking of unique characteristics of modality features. These challenges result in significant performance degradation in realistic situations where modalities are missing. To address these issues, we propose Disentangling and Generating Modality Recommender (DGMRec), a novel framework tailored for missing modality scenarios. DGMRec disentangles modality features into general and specific modality features from an information-based perspective, enabling richer representations for recommendation. Building on this, it generates missing modality features by integrating aligned features from other modalities and leveraging user modality preferences. Extensive experiments show that DGMRec consistently outperforms state-of-the-art MRSs in challenging scenarios, including missing modalities and new item settings as well as diverse missing ratios and varying levels of missing modalities. Moreover, DGMRec's generation-based approach enables cross-modal retrieval, a task inapplicable for existing MRSs, highlighting its adaptability and potential for real-world applications. Our code is available at https://github.com/ptkjw1997/DGMRec.

cs.IR↗

arXiv2025

Token-Efficient Item Representation via Images for LLM Recommender Systems

Large Language Models (LLMs) have recently emerged as a powerful backbone for recommender systems. Existing LLM-based recommender systems take two different approaches for representing items in natural language, i.e., Attribute-based Representation and Description-based Representation. In this work, we aim to address the trade-off between efficiency and effectiveness that these two approaches encounter, when representing items consumed by users. Based on our interesting observation that there is a significant information overlap between images and descriptions associated with items, we propose a novel method, Item representation for LLM-based Recommender system (I-LLMRec). Our main idea is to leverage images as an alternative to lengthy textual descriptions for representing items, aiming at reducing token usage while preserving the rich semantic information of item descriptions. Through extensive experiments, we demonstrate that I-LLMRec outperforms existing methods in both efficiency and effectiveness by leveraging images. Moreover, a further appeal of I-LLMRec is its ability to reduce sensitivity to noise in descriptions, leading to more robust recommendations.

cs.IR↗

arXiv2025

Cross, Dwell, or Pinch: Designing and Evaluating Around-Device Selection Methods for Unmodified Smartwatches

Smartwatches offer powerful features, but their small touchscreens limit the expressiveness of the input that can be achieved. To address this issue, we present, and open-source, the first sonar-based around-device input on an unmodified consumer smartwatch. We achieve this using a fine-grained, one-dimensional sonar-based finger-tracking system. In addition, we use this system to investigate the fundamental issue of how to trigger selections during around-device smartwatch input through two studies. The first examines the methods of double-crossing, dwell, and finger tap in a binary task, while the second considers a subset of these designs in a multi-target task and in the presence and absence of haptic feedback. Results showed double-crossing was optimal for binary tasks, while dwell excelled in multi-target scenarios, and haptic feedback enhanced comfort but not performance. These findings offer design insights for future around-device smartwatch interfaces that can be directly deployed on today's consumer hardware.

cs.HC↗

arXiv2025

BudsID: Mobile-Ready and Expressive Finger Identification Input for Earbuds

Wireless earbuds are an appealing platform for wearable computing on-the-go. However, their small size and out-of-view location mean they support limited different inputs. We propose finger identification input on earbuds as a novel technique to resolve these problems. This technique involves associating touches by different fingers with different responses. To enable it on earbuds, we adapted prior work on smartwatches to develop a wireless earbud featuring a magnetometer that detects fields from a magnetic ring. A first study reveals participants achieve rapid, precise earbud touches with different fingers, even while mobile (time: 0.98s, errors: 5.6%). Furthermore, touching fingers can be accurately classified (96.9%). A second study shows strong performance with a more expressive technique involving multi-finger double-taps (inter-touch time: 0.39s, errors: 2.8%) while maintaining high accuracy (94.7%). We close by exploring and evaluating the design of earbud finger identification applications and demonstrating the feasibility of our system on low-resource devices.

cs.HC↗

arXiv2025

Lost in Sequence: Do Large Language Models Understand Sequential Recommendation?

Large Language Models (LLMs) have recently emerged as promising tools for recommendation thanks to their advanced textual understanding ability and context-awareness. Despite the current practice of training and evaluating LLM-based recommendation (LLM4Rec) models under a sequential recommendation scenario, we found that whether these models understand the sequential information inherent in users' item interaction sequences has been largely overlooked. In this paper, we first demonstrate through a series of experiments that existing LLM4Rec models do not fully capture sequential information both during training and inference. Then, we propose a simple yet effective LLM-based sequential recommender, called LLM-SRec, a method that enhances the integration of sequential information into LLMs by distilling the user representations extracted from a pre-trained CF-SRec model into LLMs. Our extensive experiments show that LLM-SRec enhances LLMs' ability to understand users' item interaction sequences, ultimately leading to improved recommendation performance. Furthermore, unlike existing LLM4Rec models that require fine-tuning of LLMs, LLM-SRec achieves state-of-the-art performance by training only a few lightweight MLPs, highlighting its practicality in real-world applications. Our code is available at https://github.com/Sein-Kim/LLM-SRec.

cs.IR↗

arXiv2022

Anatomy of ultrafast quantitative magneto-acoustics in freestanding nickel thin films

We revisit the quantitative analysis of the ultrafast magneto-acoustic experiment in a freestanding nickel thin film by Kim and Bigot [1] by applying our recently proposed approach of magnetic and acoustic eigenmodes decomposition by Vernik et al. [2]. We show that the application of our modeling to the analysis of time-resolved reflectivity measurements allows for the determination of amplitudes and lifetimes of standing perpendicular acoustic phonon resonances with unprecedented accuracy. The acoustic damping is found to scale as $\proptoω^2$ for frequencies up to 80~GHz and the peak amplitudes reach $10^{-3}$. The experimentally measured magnetization dynamics for different orientations of an external magnetic field agrees well with numerical solutions of magneto-elastically driven magnon harmonic oscillators. Symmetry-based selection rules for magnon-phonon interactions predicted by our modeling approach allow for the unambiguous discrimination between spatially uniform and non-uniform modes, as confirmed by comparing the resonantly enhanced magneto-elastic dynamics simultaneously measured on opposite sides of the film. Moreover, the separation of time scales for (early) rising and (late) decreasing precession amplitudes provide access to magnetic (Gilbert) and acoustic damping parameters in a single measurement.

cond-mat.mtrl-sci↗

arXiv2022

Wide Effective Work-Function Tuning of Al/SiO$_2$/Si Junction Achieved with Graphene Interlayer at Al/SiO$_2$ Interface

The effective work-function of metal electrode is one of the major factors to determine the threshold voltage of metal/oxide/semiconductor junction. In this work, we demonstrate experimentally that the effective work-function of Aluminum (Al) electrode in Al/SiO$_2$/n-Si junction increases significantly by $\sim$1.04 eV with the graphene interlayer inserted at Al/SiO$_2$ interface. We also provide the device-physical analysis of solving Poisson equation when the flat-band voltage is applied to the junction, supporting that the wide tuning of Al effective work-function originates from the electrical dipole layer formed by the overlap of electron orbitals between Al and graphene layer. Our work suggests the feasibility of constructing the dual-metal gate CMOS circuitry just by using Al electrodes with area-specific underlying graphene interlayer.

cond-mat.mes-hall↗

arXiv2022

Accurate Ground-Truth Depth Image Generation via Overfit Training of Point Cloud Registration using Local Frame Sets

Accurate three-dimensional perception is a fundamental task in several computer vision applications. Recently, commercial RGB-depth (RGB-D) cameras have been widely adopted as single-view depth-sensing devices owing to their efficient depth-sensing abilities. However, the depth quality of most RGB-D sensors remains insufficient owing to the inherent noise from a single-view environment. Recently, several studies have focused on the single-view depth enhancement of RGB-D cameras. Recent research has proposed deep-learning-based approaches that typically train networks using high-quality supervised depth datasets, which indicates that the quality of the ground-truth (GT) depth dataset is a top-most important factor for accurate system; however, such high-quality GT datasets are difficult to obtain. In this study, we developed a novel method for high-quality GT depth generation based on an RGB-D stream dataset. First, we defined consecutive depth frames in a local spatial region as a local frame set. Then, the depth frames were aligned to a certain frame in the local frame set using an unsupervised point cloud registration scheme. The registration parameters were trained based on an overfit-training scheme, which was primarily used to construct a single GT depth image for each frame set. The final GT depth dataset was constructed using several local frame sets, and each local frame set was trained independently. The primary advantage of this study is that a high-quality GT depth dataset can be constructed under various scanning environments using only the RGB-D stream dataset. Moreover, our proposed method can be used as a new benchmark GT dataset for accurate performance evaluations. We evaluated our GT dataset on previously benchmarked GT depth datasets and demonstrated that our method is superior to state-of-the-art depth enhancement frameworks.

cs.CV↗