arXiv subjects

Seunghoon Han

Publications and source records attributed to Seunghoon Han.

16 recordsLinked to original sources

arXiv2026

Mixture-of-Translators: Translating KV Caches Across Heterogeneous Large Language Models

Heterogeneous Large Language Model (LLM) systems increasingly rely on shared contexts, retrieved evidence, and multi-agent dialogue histories, yet their internal key-value (KV) caches remain model-specific and cannot be reused across architectures. Consequently, each model must repeatedly prefill or store caches for the same context, limiting the scalability of multi-model reasoning and long-context generation. We propose Mixture-of-Translators(MoT), a cache translation framework that maps context KV caches from a source LLM into the cache space of a target LLM. Unlike prior approaches that depend on a single projection path or global shared latent space, MoT uses multiple translator modules to capture diverse source--target mappings. To further reduce residual translation error, we introduce a Context Correction Loss that aligns the replayed target trajectory with the native target trajectory. We reveal two competing failure modes in cache translation: propagated translation shift from early injection and last-state shift from late injection. MoT addresses them through translator mixtures and target-side correction. Across homogeneous and heterogeneous translations among Qwen2.5, GPT-2, and OPT models, MoT preserves downstream QA performance, including Qwen2.5-7B-scale translation with 51.0% average closed-set QA accuracy and 0.43 average extractive QA F1. In practical case studies, MoT enables quality-preserving memory reuse for multi-agent reasoning and retains 96.3% of direct-context quality in long-context cache-augmented generation, demonstrating scalable KV cache reuse across heterogeneous LLMs.

cs.CL↗

arXiv2026

MPR-CiteG: Enhancing RAG with Multi-Portfolio Retrieval and Citation-Grounded Generation

This paper presents the MPR-CiteG framework, which achieved second place in the ScienceON AI Challenge by addressing two fundamental challenges in generative AI: inefficient retrieval and the absence of source verification. We propose a dual-component system, termed MPR-CiteG, in which the Multi-Portfolio Retriever (MPR) efficiently retrieves diverse and relevant information, while the Citation-Grounded Generation (CiteG) module ensures that every generated output remains factually consistent and explicitly attributed to its source. MPR-CiteG represents a significant step toward building more trustworthy and accurate LLMs that are not only capable of generating information but also of grounding their responses in reliable evidence, thereby mitigating common issues like model hallucination. Extensive experiments on the challenge dataset validate the effectiveness and reliability of our approach. Our code is available at https://github.com/2noweyh/MPR-citeG.

cs.AI↗

arXiv2026

Hyperbolic Heterogeneous Graph Transformer

In heterogeneous graphs, we can observe complex structures such as tree-like or hierarchical structures. Recently, the hyperbolic space has been widely adopted in many studies to effectively learn these complex structures. Although these methods have demonstrated the advantages of the hyperbolic space in learning heterogeneous graphs, most existing methods still have several challenges. They rely heavily on tangent-space operations, which often lead to mapping distortions during frequent transitions. Moreover, their message-passing architectures mainly focus on local neighborhood information, making it difficult to capture global hierarchical structures and long-range dependencies between different types of nodes. To address these limitations, we propose Hyperbolic Heterogeneous Graph Transformer (HypHGT), which effectively and efficiently learns heterogeneous graph representations entirely within the hyperbolic space. Unlike previous message-passing based hyperbolic heterogeneous GNNs, HypHGT naturally captures both local and global dependencies through transformer-based architecture. Furthermore, the proposed relation-specific hyperbolic attention mechanism in HypHGT, which operates with linear time complexity, enables efficient computation while preserving the heterogeneous information across different relation types. This design allows HypHGT to effectively capture the complex structural properties and semantic information inherent in heterogeneous graphs. We conduct comprehensive experiments to evaluate the effectiveness and efficiency of HypHGT, and the results demonstrate that it consistently outperforms state-of-the-art methods in node classification task, with significantly reduced training time and memory usage.

cs.LG↗

arXiv2025

Alibaba International E-commerce Product Search Competition DILAB Team Technical Report

This study presents the multilingual e-commerce search system developed by the DILAB team, which achieved 5th place on the final leaderboard with a competitive overall score of 0.8819, demonstrating stable and high-performing results across evaluation metrics. To address challenges in multilingual query-item understanding, we designed a multi-stage pipeline integrating data refinement, lightweight preprocessing, and adaptive modeling. The data refinement stage enhanced dataset consistency and category coverage, while language tagging and noise filtering improved input quality. In the modeling phase, multiple architectures and fine-tuning strategies were explored, and hyperparameters optimized using curated validation sets to balance performance across query-category (QC) and query-item (QI) tasks. The proposed framework exhibited robustness and adaptability across languages and domains, highlighting the effectiveness of systematic data curation and iterative evaluation for multilingual search systems. The source code is available at https://github.com/2noweyh/DILAB-Alibaba-Ecommerce-Search.

cs.LG↗

arXiv2025

Ideal Optical Antimatter using Passive Lossy Materials under Complex Frequency Excitation

The original concept of left-handed material has inspired the possibility of optical antimatter, where the effect of light propagation through a medium can be completely cancelled by its complementary medium. Despite recent progress in the development of negative-index metamaterials, losses continue to be a significant barrier to realizing optical antimatter. In this work, we show that passive, lossy materials can be used to realize optical antimatter when illuminated by light at a complex frequency. We further establish that one can engineer arbitrary complex-valued permittivity and permeability in such materials. Strikingly, we show that materials with a positive index at real frequencies can act as negative-index materials under complex frequency excitation. Using our approach, we numerically demonstrate the optical antimatter functionality, as well as double focusing by an ideal perfect lens and superscattering. Our work demonstrates the power of temporally structured light in unlocking the promising opportunities of complementary media, which have until now been inhibited by material loss.

physics.optics↗

arXiv2025

Metapath-based Hyperbolic Contrastive Learning for Heterogeneous Graph Embedding

The hyperbolic space, characterized by a constant negative curvature and exponentially expanding space, aligns well with the structural properties of heterogeneous graphs. However, although heterogeneous graphs inherently possess diverse power-law structures, most hyperbolic heterogeneous graph embedding models rely on a single hyperbolic space. This approach may fail to effectively capture the diverse power-law structures within heterogeneous graphs. To address this limitation, we propose a Metapath-based Hyperbolic Contrastive Learning framework (MHCL), which uses multiple hyperbolic spaces to capture diverse complex structures within heterogeneous graphs. Specifically, by learning each hyperbolic space to describe the distribution of complex structures corresponding to each metapath, it is possible to capture semantic information effectively. Since metapath embeddings represent distinct semantic information, preserving their discriminability is important when aggregating them to obtain node representations. Therefore, we use a contrastive learning approach to optimize MHCL and improve the discriminability of metapath embeddings. In particular, our contrastive learning method minimizes the distance between embeddings of the same metapath and maximizes the distance between those of different metapaths in hyperbolic space, thereby improving the separability of metapath embeddings with distinct semantic information. We conduct comprehensive experiments to evaluate the effectiveness of MHCL. The experimental results demonstrate that MHCL outperforms state-of-the-art baselines in various graph machine learning tasks, effectively capturing the complex structures of heterogeneous graphs.

cs.LG↗

arXiv2025

Zero-Shot Industrial Anomaly Segmentation with Image-Aware Prompt Generation

Anomaly segmentation is essential for industrial quality, maintenance, and stability. Existing text-guided zero-shot anomaly segmentation models are effective but rely on fixed prompts, limiting adaptability in diverse industrial scenarios. This highlights the need for flexible, context-aware prompting strategies. We propose Image-Aware Prompt Anomaly Segmentation (IAP-AS), which enhances anomaly segmentation by generating dynamic, context-aware prompts using an image tagging model and a large language model (LLM). IAP-AS extracts object attributes from images to generate context-aware prompts, improving adaptability and generalization in dynamic and unstructured industrial environments. In our experiments, IAP-AS improves the F1-max metric by up to 10%, demonstrating superior adaptability and generalization. It provides a scalable solution for anomaly segmentation across industries

cs.CV↗

arXiv2025

CLAIMCHECK: How Grounded are LLM Critiques of Scientific Papers?

A core part of scientific peer review involves providing expert critiques that directly assess the scientific claims a paper makes. While it is now possible to automatically generate plausible (if generic) reviews, ensuring that these reviews are sound and grounded in the papers' claims remains challenging. To facilitate LLM benchmarking on these challenges, we introduce CLAIMCHECK, an annotated dataset of NeurIPS 2023 and 2024 submissions and reviews mined from OpenReview. CLAIMCHECK is richly annotated by ML experts for weakness statements in the reviews and the paper claims that they dispute, as well as fine-grained labels of the validity, objectivity, and type of the identified weaknesses. We benchmark several LLMs on three claim-centric tasks supported by CLAIMCHECK, requiring models to (1) associate weaknesses with the claims they dispute, (2) predict fine-grained labels for weaknesses and rewrite the weaknesses to enhance their specificity, and (3) verify a paper's claims with grounded reasoning. Our experiments reveal that cutting-edge LLMs, while capable of predicting weakness labels in (2), continue to underperform relative to human experts on all other tasks.

cs.CL↗

arXiv2024

Multi-Hyperbolic Space-based Heterogeneous Graph Attention Network

To leverage the complex structures within heterogeneous graphs, recent studies on heterogeneous graph embedding use a hyperbolic space, characterized by a constant negative curvature and exponentially increasing space, which aligns with the structural properties of heterogeneous graphs. However, despite heterogeneous graphs inherently possessing diverse power-law structures, most hyperbolic heterogeneous graph embedding models use a single hyperbolic space for the entire heterogeneous graph, which may not effectively capture the diverse power-law structures within the heterogeneous graph. To address this limitation, we propose Multi-hyperbolic Space-based heterogeneous Graph Attention Network (MSGAT), which uses multiple hyperbolic spaces to effectively capture diverse power-law structures within heterogeneous graphs. We conduct comprehensive experiments to evaluate the effectiveness of MSGAT. The experimental results demonstrate that MSGAT outperforms state-of-the-art baselines in various graph machine learning tasks, effectively capturing the complex structures of heterogeneous graphs.

cs.LG↗

arXiv2024

Hyperbolic Heterogeneous Graph Attention Networks

Most previous heterogeneous graph embedding models represent elements in a heterogeneous graph as vector representations in a low-dimensional Euclidean space. However, because heterogeneous graphs inherently possess complex structures, such as hierarchical or power-law structures, distortions can occur when representing them in Euclidean space. To overcome this limitation, we propose Hyperbolic Heterogeneous Graph Attention Networks (HHGAT) that learn vector representations in hyperbolic spaces with meta-path instances. We conducted experiments on three real-world heterogeneous graph datasets, demonstrating that HHGAT outperforms state-of-the-art heterogeneous graph embedding models in node classification and clustering tasks.

cs.LG↗

arXiv2022

Simulation of large-area metasurfaces with a distributed transition matrix method

Inverse design of large-area metasurfaces can potentially exploit the full parameter space that such devices offer and achieve highly efficient multifunctional flat optical elements. However, since practically useful flat optics elements are large in the linear dimension, an accurate simulation of their scattering properties is challenging. Here, we demonstrate a method to compute accurate simulations and gradients of large-area metasurfaces. Our approach relies on two key ingredients - a simulation distribution strategy that allows a linear reduction in the simulation time with number of compute (GPU) nodes and an efficient single-node computation using the Transition-matrix (T-matrix) method. We demonstrate ability to perform a distributed simulation of large-area, while accurately accounting for scatterer-scatterer interactions significantly beyond the locally periodic approximation, and efficiently compute gradients with respect to the metasurface design parameters. This scalable and accurate metasurface simulation method opens the door to gradient-based optimization of full large-area metasurfaces.

physics.optics↗

arXiv2020

Increasing efficiency of high numerical aperture metasurfaces using the grating averaging technique

One of the important advantages of optical metasurfaces over conventional diffractive optical elements is their capability to efficiently deflect light by large angles. However, metasurfaces are conventionally designed using approaches that are optimal for small deflection angles and their performance for designing high numerical aperture devices is not well quantified. Here we introduce and apply a technique for the estimation of the efficiency of high numerical aperture metasurfaces. The technique is based on a particular coherent averaging of diffraction coefficients of periodic blazed gratings and can be used to compare the performance of different metasurface designs in implementing high numerical aperture devices. Unlike optimization-based methods that rely on full-wave simulations and are only practicable in designing small metasurfaces, the gradient averaging technique allows for the design of arbitrarily large metasurfaces. Using this technique, we identify an unconventional metasurface design and experimentally demonstrate a metalens with a numerical aperture of 0.78 and a measured focusing efficiency of 77%. The grating averaging is a versatile technique applicable to many types of gradient metasurfaces, thus enabling highly efficient metasurface components and systems.

physics.optics↗

arXiv2016

Miniature optical planar camera based on a wide-angle metasurface doublet corrected for monochromatic aberrations

Optical metasurfaces are two-dimensional arrays of nano-scatterers that modify optical wavefronts at subwavelength spatial resolution. They are poised to revolutionize optics by enabling complex low-cost systems where multiple metasurfaces are lithographically stacked and integrated with electronics. For imaging applications, metasurface stacks can perform sophisticated image corrections and can be directly integrated with image sensors. Here, we demonstrate this concept with a miniature flat camera integrating a monolithic metasurface lens doublet corrected for monochromatic aberrations, and an image sensor. The doublet lens, which acts as a fisheye photographic objective, has a small $f$-number of 0.9, an angle-of-view larger than 60$^\circ\times$60$^\circ$, and operates at 850 nm wavelength with 70% focusing efficiency. The camera exhibits nearly diffraction-limited image quality, which indicates the potential of this technology in the development of optical systems for microscopy, photography, and computer vision.

physics.optics↗

arXiv2015

Gate-tunable conducting oxide metasurfaces

Metasurfaces composed of planar arrays of sub-wavelength artificial structures show promise for extraordinary light manipulation; they have yielded novel ultrathin optical components such as flat lenses, wave plates, holographic surfaces and orbital angular momentum manipulation and detection over a broad range of electromagnetic spectrum. However the optical properties of metasurfaces developed to date do not allow for versatile tunability of reflected or transmitted wave amplitude and phase after fabrication, thus limiting their use in a wide range of applications. Here, we experimentally demonstrate a gate-tunable metasurface that enables dynamic electrical control of the phase and amplitude of the plane wave reflected from the metasurface. Tunability arises from field-effect modulation of the complex refractive index of conducting oxide layers incorporated into metasurface antenna elements which are configured in a reflectarray geometry. We measure a phase shift of π and ~ 30% change in the reflectance by applying 2.5 V gate bias. Additionally, we demonstrate modulation at frequencies exceeding 10 MHz, and electrical switching of +/-1 order diffracted beams by electrical control over subgroups of metasurface elements, a basic requirement for electrically tunable beam-steering phased array metasurfaces. The proposed tunable metasurface design with high optical quality and high speed dynamic phase modulation suggests applications in next generation ultrathin optical components for imaging and sensing technologies, such as reconfigurable beam steering devices, dynamic holograms, tunable ultrathin lens, nano-projectors, and nanoscale spatial light modulators. Importantly, our design allows complete integration with electronics and hence electrical addressability of individual metasurface elements.

physics.optics↗

arXiv2012

Space-coiling metamaterials with double negativity and conical dispersion

Metamaterials are effectively homogeneous materials that display extraordinary dispersion. Negative index metamaterials, zero index metamaterials and extremely anisotropic metamaterials are just a few examples. Instead of using locally resonating elements that may cause undesirable absorption, there are huge efforts to seek alternative routes to obtain these unusual properties. Here, we experimentally demonstrate an alternative approach for constructing metamaterials with extreme dispersion by simply coiling up space with curled channels. Such a geometric approach has the advantage that the ratio between the wavelength and the lattice constant in achieving a negative or zero index is easily tunable. It allows us to construct for the first time an acoustic metamaterial with conical dispersion, leading to a clear demonstration of negative refraction from an acoustic metamaterial with airborne sound. We also design and realize a double-negative metamaterial for microwaves under the same principle.

cond-mat.mtrl-sci↗

arXiv2007

Molding the flow of light at deep sub-wavelength scale

The diffractive nature of light has limited optics and photonics to operate at scales much larger than the wavelength of light. The major challenge in scaling-down integrated photonics is how to mold the light flow below diffraction-limit in all three dimensions. A high index solid immersion lens can improve the spatial resolution by increasing the medium refractive index, but only to few times higher than in air. Photonic crystals can guide light in three dimensions, however, the guided beam width is around a wavelength. Surface plasmons has a potential to reach the sub-wavelength scales; nevertheless, it is confined in the two-dimensional interface between metals and dielectrics. Here, we present a new approach for molding the light flow at the deep sub-wavelength scale, using metamaterials with uniquely designed dispersion. We develop a design methodology for realizing sub-wavelength ray optics, and demonstrate lambda/10 width light beams flow through three-dimensional space.

physics.optics↗