arXiv subjects

Minjae Kim

Publications and source records attributed to Minjae Kim.

At least 19 recordsLinked to original sources

arXiv2026

Motif 3: Technical Report

We introduce Motif 3, a decoder-only Mixture-of-Experts language model with 314 billion total parameters and 13.2 billion activated per token. Each sparse MoE layer contains 384 routed experts, with eight selected per token. This fine-grained sparsity provides substantial expert capacity while limiting computation. Motif 3 is built around Grouped Differential Latent Attention (GDLA), which integrates grouped differential attention with the compressed key-value representation of Multi-head Latent Attention. The architecture further incorporates modified manifold-constrained hyper-connections, Expert Specific PolyNorm activations, and multi-token prediction to improve optimization stability, expert specialization, and inference efficiency. We pretrain Motif 3 on approximately 12.5 trillion tokens spanning web documents, STEM, code, mathematics, multilingual content, and domain-specialized corpora. Expert-balancing and numerical-stabilization techniques support stable training at scale, while selective MXFP8 computation and communication, memory-efficient fused kernels, and window-aware context parallelism enable training with context lengths up to 256K tokens. Our post-training pipeline combines general supervised fine-tuning, six specialist teachers trained with reinforcement learning, a software-engineering teacher trained with supervised fine-tuning, and Multi-teacher On-Policy Distillation. The resulting unified model consolidates complementary capabilities in reasoning, coding, tool use, professional work, long-context understanding, calibrated abstention, and instruction following. Across a broad evaluation suite, Motif 3 demonstrates competitive performance against leading open weight models, including strong results on long-horizon agentic tasks, mathematical reasoning, scientific knowledge, and hallucination-sensitive evaluation.

cs.AI↗

arXiv2026

The JWST Early Release Science Program for Direct Observations of Exoplanetary Systems VIII: patchy forsterite and enstatite clouds in the atmosphere of VHS 1256 b, retrieval lessons learned and outlook to the future

JWST defines a new era for the data-driven approach of retrieval modelling, which has become a cornerstone tool for the statistical inference of exoplanetary and brown dwarf properties. The Early Release Science program #1386 observations of VHS 1256 b represent a huge jump in data quality, data quantity and spectral coverage for such objects. VHS 1256 b is a young, planetary mass and extremely variable companion that populates the enigmatic L/T cohort of substellar atmospheres. In this first retrieval analysis of the full 1 - 18 micron dataset, we apply the Brewster retrieval framework to the NIRSpec and MIRI spectroscopic observations of VHS 1256 b, exploring a variety of cloud species and structures. Using Delta(BIC) we find that the data is best described by a forsterite (Mg$_{2}$SiO$_{4}$) and enstatite (MgSiO$_{3}$) cloud combination. Our analysis shows a strong preference for patchy silicate cloud coverage, which aligns with VHS 1256 b's extensive and well documented spectral variability. Our retrieval is able to place constraints on the abundances of H$_{2}$O, CO, CO$_{2}$, CH$_{4}$ as well as NH$_{3}$. We also show that the retrieved parameters are sensitive to the data used and the relative signal-to-noise ratios between data from different instruments. We conclude with the next steps for the wider retrieval community to better understand young and cloudy exoplanetary atmospheres.

astro-ph.EP↗

arXiv2026

FinVerse: Financial Time-Series Benchmark

As time-series foundation models have emerged, the need for benchmarks that can evaluate their forecasting ability in meaningful ways has become increasingly important. Existing time-series forecasting benchmarks provide useful standardized comparisons, but they often evaluate heterogeneous series with uniform error-based metrics. Strong performance under such metrics does not necessarily imply that a model's forecasts will support the best real-world decisions across domains. For example, in stock forecasting, correctly predicting whether a price will rise or fall can be more directly relevant to realized returns than minimizing point-wise forecast error alone. To this end, we introduce FinVerse, a finance-domain time-series forecasting benchmark that takes a first step toward more realistic evaluation. The released FinVerse data artifact contains 116,897 financial time series with 171.1M observations, of which 60,232 series with 17.4M observations are selected as evaluated targets based on their economic relevance to financial decisions. Unlike generic forecasting benchmarks that primarily emphasize uniform point-forecast or probabilistic accuracy, FinVerse defines 11 metric families comprising 78 evaluation metrics and assigns the most appropriate evaluation metrics to each individual time series based on its underlying economic meaning. Our analysis of 43 public time-series forecasting foundation models shows that strong performance under generic forecasting criteria does not necessarily translate into useful financial forecasts. This finding highlights the need for domain-aware benchmarks that evaluate models under objectives closer to real-world decision making.

cs.LG↗

arXiv2026

EXAONE Finance 1.0: An Attention-free Time Series Foundation Model for Financial Time Series

This technical report presents EXAONE Forecast for Finance (EXAONE Finance), a financial time series foundation model (TSFM) tailored to financial forecasting. While recent TSFMs achieve strong zero-shot performance through large-scale pretraining, they are primarily developed for general-domain time series and largely rely on self-attention backbones whose computational cost grows quadratically with sequence length and variate count. Moreover, they assume fully observed inputs and are pretrained on corpora that fail to adequately capture the unique dynamics of financial markets. These limitations hinder their applicability to finance, where long, many-channel, intermittently observed panels are common. To address these challenges, EXAONE Finance adopts an attention-free architecture, replacing self-attention with two simple yet effective linear-time operators: (1) a causal 1D convolution for temporal mixing and (2) a group-aware pooling multi-layer perceptron (MLP) for variate mixing. Furthermore, a masked-context augmentation exposes the model to contiguous missing spans during training, improving robustness to the missingness pervasive in financial markets. EXAONE Finance is pretrained on a synthetic financial corpus whose generative process is designed to reproduce the properties of financial series such as heavy tails, volatility clustering, jumps, regime shifts, and cross-asset dependence, combined with a domain-agnostic synthetic source. On FinVerse, a financial forecasting benchmark covering diverse asset classes, EXAONE Finance attains state-of-the-art performance, ranking first across all three evaluation tiers: point-forecast accuracy, cross-sectional asset ranking, and portfolio profitability.

cs.AI↗

arXiv2026

Beyond Magnitude and Shape: A Direction-Aware Loss for Time Series Forecasting

The direction of change --- whether a series will move up or down --- is often as important as its exact value in decisiondriven applications such as risk management and financial forecasting. However, most forecasting losses optimize either point magnitude or shape and frequency structure, and none explicitly targets the direction of change. In this paper, we find that MSE-trained forecasters fail on the direction of small moves. To address this, we propose CosDir, a simple yet effective direction-aware loss that aligns the difference vectors of the prediction and the target via cosine similarity. Being scale-invariant, CosDir keeps a directional gradient on small moves, re-injecting learning signal exactly where MSE neglects it. CosDir is a lightweight, plug-in term that attaches to any backbone without architectural modification. Since the best ratio for mixing the directional and magnitude terms differs across datasets, we further propose CosDir-UW, an extension that makes this ratio adaptive by learning it during training, matching a per-dataset tuned weight with no hyperparameter. We conduct over 100K experiments, demonstrating that our method consistently and significantly improves directional accuracy while preserving magnitude accuracy, and that it outperforms various loss functions. Code is available at: https://github.com/seunghan96/cosdir.

cs.LG↗

arXiv2026

ReasonCast: Towards Explainable Time Series Forecasting with Reasoning

Most time series (TS) models are specialized for a single task, either understanding (i.e., returning text answers about a TS) or generation (i.e., returning a numeric forecast). Only recently have unified models begun to handle the two within a single architecture. Even these models, however, produce the two outputs as task-separated paths and cannot predict a series and explain why that prediction arises within a single coherent response. In this paper, we argue for a task-fused model that jointly produces 1) prediction (generation) and 2) selfexplanation (understanding), thereby integrating 1) numerical TS forecasting and 2) interpretable text reasoning within a single response. To enable the systematic study of this capability, we present both a benchmark and a recipe that jointly address the two tasks. The benchmark, ReasonTS-Bench, identifies five fundamental patterns underlying TS and enables the joint evaluation of both tasks. ReasonCast, our recipe for finetuning any LLM to perform both tasks jointly, yields a model that generates a reasoning chain and a forecast together in a single autoregressive pass. Extensive experiments show that ReasonCast outperforms both LLMs and TS models on prediction accuracy while producing verifiable, causal reasoning. Code is available at: https://github.com/seunghan96/reasoncast.

cs.AI↗

arXiv2026

Unconventional and Fragile Magnetic Exciton in a van der Waals Quantum Magnet

The recently discovered magnetic exciton in the van der Waals (vdW) antiferromagnet NiPS3 exemplifies these phenomena, exhibiting several distinctive characteristics. Despite extensive investigation, much of its physics remains unresolved, with key questions about why the NiPS3 magnetic exciton is so sharp and optically bright despite the nominally spin-forbidden transition, posing significant challenges to a proper understanding and practical manipulation of the exciton. An urgent question is to what extent it is due to chemical disorder, magnetic weakening, lattice modification, or intrinsic instability of the bright exciton itself: answers to which will put stringent constraints on possible theoretical models. Here we address these questions using hydrostatic pressure as a clean, continuous, reversible, and in-situ tuning parameter. We find that the sharp photoluminescence peak is drastically suppressed by as little as 0.4 GPa and completely quenched by 1.5 GPa, with demonstrating its reversibility. Crucially, this bright-to-dark conversion occurs without magnetic, crystallographic, or electronic reconstruction despite an increase in the Neel temperature, as established by Raman, X-ray absorption, nuclear magnetic resonance spectroscopy, and first-principles many-body calculations. Our results demonstrate that the optical brightness of the magnetic exciton is independent of chemical disorder, lattice expansion, and weakening of magnetic order, indicating that a higher-order correlated mechanism governs the bright exciton. We further propose experimentally constrained microscopic scenarios involving exciton pairing, crystal-field-controlled spin-orbit mixing, and symmetry breaking, providing a framework for future tests of entangled magnetic exciton in correlated quantum magnets.

cond-mat.mtrl-sci↗

arXiv2026

Role of $p$-$d$ Hybridization on Optical Properties of Chalcopyrite Semiconductors

Designing quantum materials for coherent optical properties is a central agenda in quantum technology. Semiconductor quantum dots are an emerging approach for controlling coherent optical properties via confinement effects, tunable band gaps, and exciton binding energies, yet their inherent structural and compositional inhomogeneity degrades the coherence of the optical spectra, posing a major obstacle. We show that, for chalcopyrite semiconductors, hybridization between transition-metal $d$ and ligand $p$ electrons in the valence band is key to the coherence of the quantum dot optical spectrum. We demonstrate this using first-principles electronic-structure calculations and optical spectroscopy. The strong $p$-$d$ hybridization in CuInS$_{2}$ induces the Cu($d$) Coulomb scattering channel, giving rise to the incoherent photodoped hole carrier, while the weak $p$-$d$ hybridization in AgInS$_{2}$ induces the delocalized photodoped hole carrier having a predominant S($p$) orbital character. Our experimental results on optical spectra suggest that when the Cu ratio is enhanced in the Ag$_{1-x}$Cu$_{x}$In$_{1-y}$Ga$_{y}$S$_{2}$ quantum dot, Cu atoms at both Ag sites and defect sites experience enhanced $p$-$d$ hybridization, and a coupling begins to develop between the electrons in the quantum dot and the defect electrons at a small Cu ratio. This coupling activates Cu($d$) Coulomb scattering for photodoped holes traversing the defect sites, producing an incoherent optical response that naturally explains the long-standing absence of band-edge spectral signatures in CuIn$_{1-y}$Ga$_y$S$_2$ quantum dots. These results serve as a guideline for designing semiconductor quantum dots. To achieve a coherent optical spectrum, avoid $p$-$d$-hybridized orbital character in the photo-doped carrier.

cond-mat.mtrl-sci↗

arXiv2026

The JDISC Survey: Inner Disk Chemistry of Class I/FS Disks and Tentative Evidence for Early Pebble Drift

We present the first chemical survey of Class I and Flat-Spectrum (I/FS) disks using JWST MIRI/MRS, targeting sixteen sources in the Ophiuchus star-forming region. Through empirical line luminosity measurements and multi-component slab modeling, we characterize the molecular reservoir of these young systems and compare them to twelve Class II disks of similar stellar mass. Water, HCN, C$_2$H$_2$, and CO$_2$ are frequently detected in I/FS sources with inclinations $i < 70^{\circ}$, whereas edge-on systems show significantly suppressed emission. Compared to Class II disks, I/FS sources show suggestive---though not yet statistically significant---evidence for elevated cold water ($\sim$200\,K) mass and lower CO$_2$ excitation temperatures. Statistical analyses identify accretion luminosity as the primary correlate of molecular mass across both evolutionary stages. Once this dependence is removed, cold water and CO$_2$ masses anti-correlate with mm-dust disk radius, while hot water remains insensitive to disk size. These patterns are qualitatively consistent with pebble drift models that predict early water enrichment followed by delayed CO$_2$ delivery, suggesting an evolutionary progression from molecular-poor Class 0 sources, through water-rich Class I/FS disks, to Class II disks with reduced cold water excess. This work provides an initial evolutionary framework for disk chemistry that requires larger, multi-region samples to confirm.

astro-ph.EP↗

arXiv2026

When Summaries Distort Decisions: Information Fidelity in LLM-Compressed Financial Analysis

Financial decision-makers face more information than they can directly inspect, making context compression necessary. Yet when large language models (LLMs) compress financial source material, they can alter the investment judgment supported by the original source. We frame this problem as information fidelity: compression loses fidelity when it changes the decision induced by the source. In agentic systems, such losses may recur across intermediate steps and amplify throughout the decision process. Across financial filings and earnings-call transcripts, we find that LLM-based compression can produce fluent and factually plausible compressed contexts that nevertheless alter downstream decisions. We analyze two diagnostic patterns associated with fidelity loss: decontextualization, where salient evidence is retained but separated from the caveats and contextual qualifiers needed for correct interpretation, and model dependency, where different compressors expose different views of the same source. We then propose Agentic Context Compression, which generates multiple candidate compressions and audits their disagreements against the original source. Our results suggest that financial compression should be evaluated not only by efficiency or factuality, but also by its ability to preserve decision-relevant context.

cs.AI↗

arXiv2026

PDE-constrained optimization for virtual sensing in structural dynamics: Full-field displacement and force recovery from sparse sensors

Virtual sensing -- recovering full-field structural response from sparse sensor measurements -- is a fundamental challenge in structural health monitoring (SHM). Fracture assessment, fatigue evaluation and remaining useful life (RUL) prognosis all take the applied load as their input, and that load frequently acts on surfaces that cannot be instrumented. An improved load estimate is therefore of direct practical value for SHM and for prognostics and health management (PHM). The mode-based approach reconstructs the displacement field accurately from a small number of sensors and has been applied successfully for that purpose. The load inferred from this reconstructed displacement is far less reliable, because the residual displacement error is amplified when it is mapped back through the dynamic stiffness. This study therefore formulates virtual sensing as a PDE-constrained optimization (PDE-CO) problem in which the elastodynamic equation is an equality constraint and the applied load is the optimization variable, so that displacement and load are estimated together rather than in sequence. To manage the computational cost, the framework separates offline finite element model preparation from online reconstruction. The assembled matrices are reused during online optimization, and the forward and gradient evaluations can use GPU-based parallel computation. The numerical comparisons support improved noise robustness of PDE-CO displacement reconstruction relative to the mode-based approach within the tested settings. These results provide a theoretical and computational foundation for the development of virtual sensing for SHM and PHM.

math-ph↗

arXiv2026

MemRefine: LLM-Guided Compression for Long-Term Agent Memory

Large language model (LLM) agents are increasingly expected to operate over long-term interactions, where information from past dialogues must be preserved and recalled to support future tasks. However, as interactions accumulate, the memory store grows without bound and fills with redundant entries that inflate storage cost and degrade retrieval by crowding out the most useful evidence. Furthermore, this is especially limiting on resource-constrained platforms with hard memory budgets, motivating us to formulate storage-budgeted memory management, the task of keeping an already constructed memory store within a fixed budget while preserving information useful for future interactions. To this end, we then propose MemRefine, an LLM-guided framework that, since surface similarity poorly reflects factual value, uses similarity only to propose candidate pairs and defers delete, merge, and preserve decisions to an LLM judge based on factual content, iterating until the budget is met. Across multiple memory frameworks and long-term conversation benchmarks, MemRefine consistently meets target budgets while preserving downstream performance and outperforming rule-based baselines under tight budgets.

cs.CL↗

arXiv2026

FoRA: Fisher-orthogonal Rank Adaptation for Parameter-Efficient Fine-Tuning

Parameter-efficient fine-tuning(PEFT) has largely focused on LoRA and its accuracy-oriented variants, leaving the original goal of reducing trainable parameters has receivedcomparatively little attention. We introduce FoRA, which revisits this goal by reducing the number of adapted layers rather than adapter rank. FoRA selects task-informative layers via a single-pass diagonal Fisher score (under 1% of training cost) and trains the LoRA down-projection at selected layers on the Stiefel manifold, preserving column orthonormality and effective rank. FoRA consistently outperforms LoRA and DoRA at half their parameter budget, and falls within 0.7-0.8 accuracy points of AdaLoRA at one-quarter its parameter count, across five LLaMA-family backbones. Cross-architecture experiments on twelve backbones from the LLaMA, Qwen3, and Gemma families confirm consistent gains from 270M to 32B parameters. The two components combine super-additively: Fisher selection alone matches rank reduction at the same budget, while the Stiefel constraint provides the decisive additional gain.

cs.CL↗

arXiv2026

WorldKV: Efficient World Memory with World Retrieval and Compression

Autoregressive video diffusion models have enabled real-time, action-conditioned world generation. However, sustaining a persistent world, where revisiting a previously seen viewpoint yields consistent content, remains an open problem. Full KV-cache attention preserves this consistency but breaks real-time constraints: memory footprint and attention cost grow linearly with rollout length. Sliding window inference restores throughput but discards long-term consistency. We propose WorldKV, a training-free framework with two components: World Retrieval and World Compression. World Retrieval stores evicted KV-cache chunks in GPU/CPU memory and selectively retrieves scene-relevant chunks via camera/ action correspondence, inserting them back into the native attention window without re-encoding. World Compression prunes redundant tokens within each chunk via key-key similarity to an anchor frame, halving per-chunk storage to fit 2x more history under a fixed budget. On Matrix-Game-2.0 and LingBot- World-Fast, WorldKV matches or exceeds full-KV memory fidelity at roughly 2x the throughput, and is competitive with memory-trained baselines without any fine-tuning. Project Page: https://cvlab-kaist.github.io/WorldKV/

cs.CV↗

arXiv2026

AdaTKG: Adaptive Memory for Temporal Knowledge Graph Reasoning

Temporal knowledge graphs (TKGs) represent time-stamped relational facts and support a wide range of reasoning tasks over evolving events. However, existing methods produce entity representations that are static at the entity level, in that each representation is a function of learned parameters only and retains no trace of the interactions in which the entity has participated. In this paper, we depart from this static view and propose that each entity be modeled as an adaptive process whose representation is refined every time the entity participates in a fact. To this end, we propose AdaTKG, which maintains a per-entity memory that is updated with every observed interaction, with the memory accumulating online and predictions improving as more interactions arrive. Specifically, we instantiate the memory update as a learnable exponential moving average governed by a single shared scalar instead of using learnable parameters for each entity, enabling AdaTKG to handle entities unseen during training. Extensive experiments confirm consistent gains over TKG baselines, demonstrating the effectiveness of adaptive memory. Code is available at: https://github.com/seunghan96/AdaTKG

cs.AI↗

arXiv2026

FinSTaR: Towards Financial Reasoning with Time Series Reasoning Models

Time series (TS) reasoning models (TSRMs) have shown promising capabilities in general domains, yet they consistently fail on financial domain, which exhibit unique characteristics. We propose a general 2 x 2 capability taxonomy for TSRMs by crossing 1) single-entity vs. multi-entity analysis with 2) assessment of the current state vs. prediction of future behavior. We instantiate this taxonomy in the financial domain---where the distinction between deterministic assessment and stochastic prediction is particularly critical---as ten financial reasoning tasks, forming the FinTSRBench benchmark based on S&P stocks. To this end, we propose FinSTaR (Financial Time Series Thinking and Reasoning), trained on FinTSR-Bench with distinct chain-of-thought (CoT) strategies tailored to each category. For assessment, which is deterministic, we employ Compute-in-CoT, a programmatic CoT that enables models to derive answers directly from raw prices. For prediction, which is inherently stochastic, we adopt Scenario-Aware CoT, which generates diverse scenarios before making a judgment, mirroring how financial analysts reason under uncertainty. FinSTaR achieves 78.9% average accuracy on FinTSRBench, substantially outperforming LLM and TSRM baselines. Furthermore, we show that the four capability categories are complementary and mutually reinforcing through joint training. Code is publicly available at: https://github.com/seunghan96/FinSTaR.

cs.AI↗

arXiv2026

Motif-Video 2B: Technical Report

Training strong video generation models usually requires massive datasets, large parameter counts, and substantial compute. In this work, we ask whether strong text-to-video quality is possible at a much smaller budget: fewer than 10M clips and less than 100,000 H200 GPU hours. Our core claim is that part of the answer lies in how model capacity is organized, not only in how much of it is used. In video generation, prompt alignment, temporal consistency, and fine-detail recovery can interfere with one another when they are handled through the same pathway. Motif-Video 2B addresses this by separating these roles architecturally, rather than relying on scale alone. The model combines two key ideas. First, Shared Cross-Attention strengthens text control when video token sequences become long. Second, a three-part backbone separates early fusion, joint representation learning, and detail refinement. To make this design effective under a limited compute budget, we pair it with an efficient training recipe based on dynamic token routing and early-phase feature alignment to a frozen pretrained video encoder. Our analysis shows that later blocks develop clearer cross-frame attention structure than standard single-stream baselines. On VBench, Motif-Video~2B reaches 83.76\%, surpassing Wan2.1 14B while using 7$\times$ fewer parameters and substantially less training data. These results suggest that careful architectural specialization, combined with an efficiency-oriented training recipe, can narrow or exceed the quality gap typically associated with much larger video models.

cs.CV↗

arXiv2026

Rethinking Multimodal Fusion for Time Series: Text Modalities Need Constrained Fusion

Recent advances in multimodal learning have motivated the integration of auxiliary modalities such as text or vision into time series (TS) forecasting. However, most existing methods provide limited gains, often improving performance only in specific datasets or relying on architecture-specific designs that limit generalization. In this paper, we show that multimodal models with naive fusion strategies (e.g., simple addition or concatenation) often underperform unimodal TS models, which we attribute to the uncontrolled integration of auxiliary modalities which may introduce irrelevant information. Motivated by this observation, we explore various constrained fusion methods designed to control such integration and find that they consistently outperform naive fusion methods. Furthermore, we propose Controlled Fusion Adapter (CFA), a simple plug-in method that enables controlled cross-modal interactions without modifying the TS backbone, integrating only relevant textual information aligned with TS dynamics. CFA employs low rank adapters to filter irrelevant textual information before fusing it into temporal representations. We conduct over 20K experiments across various datasets and TS/text models, demonstrating the effectiveness of the constrained fusion methods. Code is available at: https://github.com/seunghan96/cfa.

cs.LG↗