Searcharxiv⌕ Search

arXiv subjects

Sangmin Lee

Publications and source records attributed to Sangmin Lee.

At least 19 recordsLinked to original sources
arXiv2026

Persistent Teacher Anchoring for Tool-Using Agents

Distillation is common in LLM post-training, where on-policy knowledge distillation (OPKD) uses student-generated trajectories to prepare the student for downstream RL. At each state, the student matches a next-token distribution supplied by the teacher. As the rollout enters states the teacher would not visit, the teacher-student distribution gap can accumulate. In tool use, this gap becomes consequential because student-written calls execute before supervision and their observations shape later prefixes. Proposer-verifier generation addresses this drift by letting the teacher decide which student-proposed text is retained during generation. Existing formulations govern text but leave tool execution outside their scope. We propose Persistent Teacher Anchoring (PTA), a student-induced but teacher-committed rollout construction. PTA retains chunk-level verification and adds turn-level commitment, allowing a call to reach the environment only after the teacher has verified the entire turn. Treating verified chunks as atomic generation units, we introduce persistent lookahead, which fills idle rollout capacity by advancing future samples and carrying unfinished ones across student updates under the fixed verifier. Across Search-R1-style retrieval and DeepEyes-style perception RL, applying PTA before downstream RL improves macro best@4 by 2.5 and 2.8 points over OPKD under the same downstream RL budget, while lookahead improves throughput by 24%.

cs.LG↗
arXiv2026

Verification-Aware Training for Speculative Decoding

Speculative decoding accelerates large language model inference by using a draft model to generate candidate tokens, which are verified by the target model in a single forward pass. Verification proceeds sequentially and discards every position from the first rejection onward, yet existing draft training relies on token-level imitation of the target with a fixed per-position weighting that reflects neither property. We introduce Verification-Aware Training (VAT), a plug-in framework that simulates verification at every training step and turns the resulting accept and reject patterns into supervision. VAT consists of two components: (i) a verification head, a lightweight jointly trained binary classifier that supervises the draft model on whether each position survives sequential verification; (ii) verification-adaptive weighting, which replaces the fixed weighting schedule by keeping full weight up to each sample's first rejection point and re-anchoring the decay to start there. VAT modifies only the training objective, so it can be layered on top of existing methods without changing the draft architecture, the target model, or the inference procedure. Applied to EAGLE-3 and DFlash on Qwen3-4B, Qwen3-8B, and LLaMA-3.1-8B, VAT improves average acceptance length by up to 11.4% and wall-clock speedup by up to 8.7%, with consistent gains across math, code, and chat benchmarks. Code will be available at https://github.com/naver-ai/vat

cs.CL↗
arXiv2026

Rationale-Guided Learning for Multimodal Emotion Recognition

Multimodal emotion recognition in conversation (MERC) requires understanding complex interactions between verbal and non-verbal cues. However, most existing approaches fundamentally treat this as a direct input-output (multimodal cues-emotion labels) mapping problem, overlooking the causal reasoning that humans use when interpreting emotions. We propose rationale-guided learning (RGL), a novel framework that transforms MERC into a cognitively-inspired reasoning task. Based on dual-process theory, we decompose emotional reasoning into three facets: Intuitive (immediate perception, System 1), Contextual (situational analysis, System 2), and Integrative (synthesis of both). We leverage an MLLM offline to generate structured rationales, which are encoded as memories to guide model training via aligning internal representations with human-like reasoning patterns. Our final model operates without any MLLM overheads at inference time. Experimental results show that RGL achieves state-of-the-art performance on the IEMOCAP and MELD benchmarks. Further, for interpretation, we demonstrate that the model's internal features effectively retrieve semantically correct rationales for unseen test samples, validating its rationale reasoning capabilities.

cs.AI↗
arXiv2026

MoLGE: Mixture of Language Group Experts for Efficient Scaling of Massively Multilingual Speech Recognition

Massively multilingual automatic speech recognition (ASR) models covering hundreds of languages must maintain robust performance across diverse linguistic and acoustic conditions. However, these models often encounter the curse of multilinguality, where model capacity is diluted across languages. To address this challenge, we propose Mixture of Language Group Experts (MoLGE), built upon speech self-supervised models (S3Ms). MoLGE assigns dedicated expert modules to clusters of similar languages, reducing the number of required submodules compared to conventional language-specific Mixture-of-Experts (MoE) schemes. It further integrates a hierarchical Low-Rank Adaptation (LoRA) strategy into the disentangled acoustic and linguistic components of the S3M architecture, enabling efficient modeling of language-specific characteristics while maintaining parameter efficiency. Further, we investigate the impact of language grouping strategies based on both linguistic and data-driven criteria on overall performance, providing an interpretable perspective on how language structure influences scalability in multilingual speech systems. In experiments, we evaluate MoLGE on a multilingual benchmark encompassing 495 languages. Results demonstrate that MoLGE consistently outperforms dense multilingual baselines with a minimal increase in trainable parameters. Notably, these language grouping strategies yield substantial improvements for both phonetic and orthographic aspects of ASR modeling. Our findings suggest that structured language specialization provides an effective pathway for massively scaling language coverage of multilingual ASR.

cs.CL↗
arXiv2026

SubSplat: High-Resolution Pixel-aligned 3DGS via Sub-pixel Gaussian Reparameterization

Pixel-aligned Gaussian splatting enables efficient and generalizable novel-view synthesis. However, high-resolution rendering faces a critical trade-off where increasing input resolution improves detail at the expense of quadratically rising network computational cost. Conversely, maintaining low-resolution inputs stabilizes this cost but results in insufficient Gaussian density and artifacts. To address this, we propose SubSplat, which introduces Sub-pixel Gaussian Reparameterizer(SPGR) to subdivide primary Gaussians into fine-grained primitives, restoring structural density directly from low-resolution features. We further enhance the reparameterization quality through feature aggregation, which effectively captures high-frequency details across multiple views. Experiments on RealEstate10K and ACID demonstrate that SubSplat achieves high-fidelity rendering with superior efficiency. Our results validate that the proposed framework successfully resolves the trade-off between reparameterization fidelity and network computational cost inherent in pixel-aligned Gaussian Splatting.

cs.CV↗
arXiv2026

AEGIS: Awareness-Enhanced Guidance for Iterative Safeguard

Span-level rationales are often assumed to improve controllability in text detoxification, but it remains unclear when such guidance helps and when it introduces trade-offs. We present Awareness-Enhanced Guidance for Iterative Safeguard (AEGIS) as an exploratory framework for studying span-guided multilingual detoxification across English, Mandarin Chinese, and Korean. AEGIS combines span-level detector outputs with frozen generator backbones, allowing harmful spans, intensity labels, and target attributes to be provided as structured guidance during rewriting. Rather than claiming state-of-the-art detoxification performance, we analyze how span guidance affects the balance between toxicity reduction and meaning preservation across generator families, model scales, and languages. Our results suggest that span-guided detoxification is conditionally useful: explicit rationales change the trade-off between toxicity reduction and meaning preservation, but their effects depend strongly on the generator backbone and the linguistic context. These findings highlight both the promise and the limitations of span-level control signals for multilingual detoxification.

cs.CL↗
arXiv2026

UR-BERT: Scaling Text Encoders for Massively Multilingual TTS Through Universal Romanization and Speech Token Prediction

We propose UR-BERT, a Romanized transcription-based text-to-speech (TTS) encoder for massively multilingual TTS systems. Conventional grapheme-to-phoneme (G2P)-based approaches are limited to around 100 languages due to the availability of reliable G2P resources. In contrast, UR-BERT scales to 495 languages by unifying diverse writing systems into a shared Romanization representation. To further enhance phonetic fidelity and text-speech alignment, we introduce a speech token prediction objective during training, which encourages the encoder to learn speech-aware phonetic representations in a data-efficient manner. Experiments show that TTS systems built on UR-BERT consistently outperform recent text encoder baselines across a wide range of languages and resource conditions, and demonstrate strong generalization to unseen languages.

cs.CL↗
arXiv2026

The Diagrammar of Quantum Magnusian

The logarithm of the time-evolution operator has been termed Magnusian, on account of the fact that its expansion describes the Magnus series. The diagrammatic expansion and computation of the classical Magnusian have been completely established in terms of tree graphs and their Hopf algebra. Recent works initiated extensions into quantum field theory, revealing general structures of loop expansions while finding intriguing relations between different diagrams. In this work, we advance the loop expansion further by providing an efficient diagrammatic algorithm to calculate the weight factor of each graph in the quantum Magnusian, known as the Murua coefficient. This is achieved by incorporating two complementary perspectives on the Magnusian at the same time: the color basis and the black-and-white basis. We extract the Murua coefficients from the Magnus series by utilizing these two bases while implementing an exponentiated Wick contraction. In turn, we identify the loop-level extension of Murua's recursive formula. Eventually, we establish a set of edge-contraction rules which facilitate a direct recursive computation of the Murua coefficients at the purely diagrammatic level, without referencing or directly manipulating the underlying Magnus expansion. This shows that the matrix elements of the quantum Magnusian can be computed from graph manipulations alone.

hep-th↗
arXiv2026

GRASP: Learning to Ground Social Reasoning in Multi-Person Non-Verbal Interactions

Understanding social interactions requires reasoning over subtle non-verbal cues, yet current multimodal large language models (MLLMs) often fail to identify who interacts with whom in multi-person videos. We introduce GRASP, a large-scale social reasoning dataset that connects high-level social QA with fine-grained gaze and deictic gesture events. GRASP contains 290K question--answer pairs over 46K videos totaling 749 hours, organized by a 16-category taxonomy spanning gaze, gesture, and joint gaze--gesture reasoning, together with GRASP-Bench for evaluation. Unlike prior resources that focus on either isolated cues or high-level social QA, GRASP builds questions from identity-consistent gaze trajectories, deictic gestures, and their joint compositions into social events. Moreover, we propose Social Grounding Reward (SGR), a learning signal that uses these social events to encourage models to reason about the participants involved in each interaction. Experiments show that SGR improves performance on GRASP-Bench while maintaining zero-shot performance on related social video QA benchmarks.

cs.CV↗
arXiv2026

On the integrability of root-Kerr probe dynamics

In the background of a Kerr-Newman black hole, the motion of a scalar particle is integrable by virtue of an extra conserved charge known as Carter charge. When the particle is endowed with spin, it is known that another conserved charge, the R\"udiger charge, maintains the integrability at least at low orders in the spin magnitude. We explore the extent of this integrability in a simpler model where both the source and the probe are root-Kerr particles, the non-gravitating limit of the Kerr-Newman black hole. At the leading order in the probe charge, the integrability holds to all orders in the spin magnitude if the interaction vertices of the probe are dictated by the Newman-Janis shift. At the second order in the probe charge, the integrability can be extended to the spin-squared order but begins to fail at the spin-cubic order. An argument based on asymptotic conservation suggests that it is impossible to restore the conservation at the spin-cubic order by a further deformation of the probe action. We compare our results with related observations for Kerr black holes with gravitational interactions.

hep-th↗
arXiv2026

From Adaptation to Generalization: Adaptive Visual Prompting for Medical Image Segmentation

Visual prompting has emerged as a powerful method for adapting pre-trained models to new domains without updating model parameters. However, existing prompting methods typically optimize a single prompt per domain and apply it uniformly to all inputs, limiting their ability to generalize under intra and inter-domain variability, which is especially critical in the medical field. To address this, we propose APEX, an Adaptive Prompt EXtraction framework that retrieves input-specific prompts from a learnable prompt memory. The memory stores diverse, domain-discriminative prompt representations and is queried via domain features extracted from the Fourier spectrum. To learn robust and discriminative domain features, we introduce a novel Low-Frequency Feature Contrastive (LFC) learning framework that clusters representations from the same domain while separating those from different domains. Extensive experiments on two medical segmentation tasks demonstrate that APEX significantly improves generalization across both seen and unseen domains. Furthermore, it complements any existing backbones and consistently enhances performance, confirming its effectiveness as a plug-and-play prompting solution in medical fields. The code is available at https://github.com/cetinkayaevren/apex/

cs.CV↗
arXiv2026

Universality in Relativistic Spinning Particle Models

We establish an equivalence between massive spinning particle models in four spacetime dimensions coupled to electromagnetism or gravity, within the spin-magnitude-preserving sector. Four representative models in the literature are shown to describe exactly the same physics in their free and interacting theories: vector oscillator, spinor oscillator, spherical top, and massive twistor. The Bargmann-Michel-Telegdi (BMT) and quadrupolar Mathisson-Papapetrou-Dixon (QMPD) equations are derived in a model-independent fashion. This universal framework allows for incorporating higher spin multipole interactions as well. We establish the rigorous construction of the interacting theory of the spherical top model with emphasis on spin gauge invariance. Applications to black hole physics, conserved charges, and post-Newtonian or post-Minkowskian frameworks are discussed.

hep-th↗
arXiv2026

Interfacial Potential Transduction for Diagnostics

A major barrier to decentralized, near-patient diagnostics is the lack of a signal transduction modality that is both analytically precise and accessible at the point of care. Optical readouts remain instrument-dependent and difficult to miniaturize, while compact electrochemical readouts are prone to matrix-derived signal distortion, limiting their biomarker coverage in real clinical settings. Here, we define interfacial potential transduction as a standardized electrical modality for portable, clinical-grade diagnostics across diverse assay formats. A mechanistic framework identifying key sample matrix parameters within the interfacial potentials transduction system enables control of biofluid-derived interference, and is demonstrated in a widely accessible lateral flow immunoassay format through quantitative detection of estradiol, progesterone, and luteinizing hormone in human plasma with high correlation (r2 > 0.97) to clinical analyzers. Broader applicability across representative diagnostic sectors is further demonstrated through exceptional performance including glucose quantification for biochemical analysis with limit of detection (LOD) of 0.92 ug/dL, HIV p24 capsid protein under an immunomagnetic separation workflow (LOD = 44.8 fg/mL), and hepatitis B virus detection within 5 min via loop-mediated isothermal amplification for molecular diagnostics. Together, these results establish interfacial potentials transduction as a unified diagnostic paradigm for near-patient deployment beyond optical and electrochemical approaches.

q-bio.BM↗
arXiv2026

KD-EKF: Knowledge-Distilled Adaptive Covariance EKF for Robust UWB/PDR Indoor Localization

Ultra-wideband (UWB) indoor localization provides centimeter-level accuracy and low latency, but its measurement reliability degrades severely under Non-Line-of-Sight (NLOS) conditions, leading to meter-scale ranging errors and inconsistent uncertainty characteristics. Inertial Measurement Unit (IMU)-based Pedestrian Dead Reckoning (PDR) complements UWB by providing infrastructure-free motion estimation; however, its error accumulates nonlinearly over time due to bias and noise propagation. Fusion methods based on Extended Kalman Filters (EKF) and Particle Filters (PF) can improve average localization accuracy through probabilistic state estimation. However, these approaches typically rely on manually tuned measurement covariances. Such fixed or heuristically tuned parameters are hard to sustain across varying indoor layouts, NLOS ratios, and motion patterns, leading to limited robustness and poor generalization of measurement uncertainty modeling in heterogeneous environments. To address this limitation, this work proposes an adaptive measurement covariance scaling framework in which reliability cues are learned from historical UWB/PDR trajectories. A large teacher model is employed offline to generate temporally consistent next-position predictions from structured UWB/PDR sequences, and this behavior is distilled into a lightweight student model suitable for real-time deployment. The student model continuously regulates EKF measurement covariances based on prediction residuals, enabling environment-aware fusion without manual re-tuning. Experimental results demonstrate that the proposed KD-EKF framework significantly reduces localization error, suppresses error spikes during Line-of-Sight (LOS)/NLOS transitions, and mitigates long-term drift compared to fixed-parameter EKF, thereby improving measurement robustness across diverse indoor environments.

eess.SP↗
arXiv2026

MuCo: Multi-turn Contrastive Learning for Multimodal Embedding Model

Universal Multimodal embedding models built on Multimodal Large Language Models (MLLMs) have traditionally employed contrastive learning, which aligns representations of query-target pairs across different modalities. Yet, despite its empirical success, they are primarily built on a "single-turn" formulation where each query-target pair is treated as an independent data point. This paradigm leads to computational inefficiency when scaling, as it requires a separate forward pass for each pair and overlooks potential contextual relationships between multiple queries that can relate to the same context. In this work, we introduce Multi-Turn Contrastive Learning (MuCo), a dialogue-inspired framework that revisits this process. MuCo leverages the conversational nature of MLLMs to process multiple, related query-target pairs associated with a single image within a single forward pass. This allows us to extract a set of multiple query and target embeddings simultaneously, conditioned on a shared context representation, amplifying the effective batch size and overall training efficiency. Experiments exhibit MuCo with a newly curated 5M multimodal multi-turn dataset (M3T), which yields state-of-the-art retrieval performance on MMEB and M-BEIR benchmarks, while markedly enhancing both training efficiency and representation coherence across modalities. Code and M3T are available at https://github.com/naver-ai/muco

cs.IR↗
arXiv2025

First-principles evidence for conventional superconductivity in a quasicrystal approximant

Quasicrystals (QCs) host long-range order without translational symmetry, a regime in which the very foundations of BCS theory are not straightforwardly applicable, yet experiments on QCs and their approximant crystals (ACs) point to conventional, $s$-wave, electron-phonon coupled superconductivity. Here we test the predictive power of the electron-phonon framework in a representative decagonal AC from first principles. Using state-of-the-art \textit{ab initio} methods, we compute the superconducting properties of the recently discovered AC Al$_{13}$Os$_4$ and quantitatively reproduce its bulk $T_\text{c}$. This constitutes, to our knowledge, the first \textit{ab initio} determination of $T_\text{c}$ for an AC and establishes that the electron-phonon framework is predictive in these systems as well. Using the generalized quasichemical approximation for alloy modeling in the decagonal Al--Os family, we predict tunable superconductivity in Al$_{13}$Os$_{4-x}$Re$_x$ and Al$_{13}$Os$_{4-x}$Ir$_x$; in particular, Al$_{13}$Re$_4$ is dynamically stable and estimated to have a $T_\text{c}$ about 30% above Al$_{13}$Os$_4$. Finally, we discuss the role of ACs as high-fidelity proxies for their parent QCs. Although long-range quasiperiodicity may introduce subtle electronic features, our findings indicate that the key ingredients for superconductivity are already encoded in the local structural motifs preserved by the AC. This places the Al--Os and Al--Re families among the most promising candidates for the highest-$T_\text{c}$ quasicrystalline superconductivity.

cond-mat.supr-con↗
arXiv2025

K-DRIFT: Unveiling New Imagery of the Hidden Universe

Low-surface-brightness (LSB) structures play a crucial role in understanding galaxy evolution by providing significant insights into galaxy interactions, the histories of mass assembly, and the distribution of dark matter. Nevertheless, their inherently faint nature, coupled with observational difficulties such as stray light interference and variations in the sky background, has significantly impeded comprehensive studies of LSB features. The KASI Deep Rolling Imaging Fast Telescope (K-DRIFT) project aims to address these observational challenges by developing off-axis freeform three-mirror telescopes and observational strategies specifically designed for LSB imaging surveys. The first generation of the K-DRIFT (K-DRIFT G1) has been successfully completed, and the forthcoming survey, scheduled to commence shortly, is expected to yield novel insights into the LSB universe. This paper outlines the scientific motivations of the project, discusses the technical challenges encountered, highlights the innovative solutions devised, and describes the future trajectory of the K-DRIFT.

astro-ph.GA↗
arXiv2025

LAMA-UT: Language Agnostic Multilingual ASR through Orthography Unification and Language-Specific Transliteration

Building a universal multilingual automatic speech recognition (ASR) model that performs equitably across languages has long been a challenge due to its inherent difficulties. To address this task we introduce a Language-Agnostic Multilingual ASR pipeline through orthography Unification and language-specific Transliteration (LAMA-UT). LAMA-UT operates without any language-specific modules while matching the performance of state-of-the-art models trained on a minimal amount of data. Our pipeline consists of two key steps. First, we utilize a universal transcription generator to unify orthographic features into Romanized form and capture common phonetic characteristics across diverse languages. Second, we utilize a universal converter to transform these universal transcriptions into language-specific ones. In experiments, we demonstrate the effectiveness of our proposed method leveraging universal transcriptions for massively multilingual ASR. Our pipeline achieves a relative error reduction rate of 45% when compared to Whisper and performs comparably to MMS, despite being trained on only 0.1% of Whisper's training data. Furthermore, our pipeline does not rely on any language-specific modules. However, it performs on par with zero-shot ASR approaches which utilize additional language-specific lexicons and language models. We expect this framework to serve as a cornerstone for flexible multilingual ASR systems that are generalizable even to unseen languages.

cs.CL↗