Searcharxiv⌕ Search

arXiv subjects

Jaeyoung Do

Publications and source records attributed to Jaeyoung Do.

At least 19 recordsLinked to original sources
arXiv2026

Beyond Retraining-Free MoE Compression: A Cost-Normalized Study of Post-Compression Adjustment

Retraining-free MoE compression reduces deployment memory by pruning or merging experts, but often treats the compressed checkpoint as the final artifact. We argue that this view is incomplete: compressed MoE checkpoints are better understood as compressed initializations that benefit from a tiny post-compression adjustment stage. Across two MoE LLM backbones, four pruning/merging methods, three expert-retention ratios, and 28 benchmarks, we compare LM fine-tuning and teacher-based KD under matched small-data budgets and measured GPU costs. Using only 3,000 C4 examples and a single epoch of adjustment, Full FT recovers 37.3% of the original-to-compressed performance gap on average. Moreover, LM fine-tuning is more cost-effective than standard token-level KD, and full-parameter adjustment gives the strongest cost--recovery trade-off among the tested scopes. These results suggest that retraining-free compression should be paired with small post-compression adjustment to recover a substantial portion of the performance lost during compression.

cs.LG↗
arXiv2026

Unified Static-Dynamic Pruning for Efficient LLM Inference

The increasing deployment of large language models (LLMs) has magnified the computational and memory bottlenecks of autoregressive decoding, where low compute intensity and bandwidth-bound kernels dominate inference cost. Weight pruning offers a promising remedy, but existing methods remain confined to either static pruning (SP), which permanently removes redundant weights but lacks adaptivity, or dynamic pruning (DP), which adapts to input sparsity but introduces runtime irregularity. This paper presents SPDP, a unified sparse-inference framework that integrates unstructured SP with input-adaptive DP for efficient LLM inference on GPUs. SPDP co-designs a new Tiled-Column-wise Bitmap Compressed (Tiled-CBC) format and two complementary GPU kernels: (1) a CUDA-core spMspV kernel featuring Hybrid Activation-aware Dynamic Shared-Memory Bitmap Decoding (HAD-SMBD) for fine-grained, runtime activation skipping, and (2) a Tensor-Core SpMM kernel optimized for prefill computation. This joint format-kernel design harmonizes static and dynamic sparsity, maintaining bandwidth-efficient memory access and high compute intensity under both phases of LLM inference. Comprehensive evaluations on inference-optimized GPUs demonstrate that SPDP achieves 1.24x-1.37x average speedup (up to 2.51x) over state-of-the-art sparse frameworks such as SpInfer, while matching perplexity with up to 25% higher sparsity. SPDP advances the inference efficiency-quality Pareto frontier, showing that unified static-dynamic pruning can deliver substantial throughput and performance-per-watt improvements in large-scale LLM serving.

cs.DC↗
arXiv2026

Exploiting Structural Properties for Efficient Constraint-Aware HNSW Hyperparameter Tuning

Vector databases (VectorDBs) are a core component of modern retrieval systems, including Retrieval-Augmented Generation (RAG), where efficient Approximate Nearest Neighbor Search (ANNS) is critical. Among ANNS algorithms, Hierarchical Navigable Small World (HNSW) graphs are widely adopted for their strong recalllatency trade-off. However, configuring HNSW remains challenging: its hyperparameters jointly affect search quality, latency, build time, and index size in nonlinear ways, while production deployments impose strict resource and tuning-time constraints.We study HNSW hyperparameter tuning from a systems perspective and show that its configuration space exhibits strong structural regularities. Specifically, we identify monotonic, dominant unimodal, and separable relationships among search-time and construction-time parameters, which induce feasibility boundaries under performance and resource constraints. Building on this insight, we propose CHAT, a constraint-aware tuning framework for HNSW. Unlike generic black-box optimizers, CHAT exploits HNSW-specific structure to perform deterministic, sample-efficient search and prune resource-infeasible configurations before full index construction. Across multiple datasets and HNSW-based vector search engines, CHAT identifies configurations that maximize recall or throughput while satisfying constraints on accuracy, latency, build time, index size, and tuning budget. Compared to strong baselines, CHAT achieves up to 45% higher throughput or 11% higher recall, and converges up to 44x faster. These results show that principled, structure-aware tuning enables efficient and robust HNSW deployment beyond generic black-box optimization.

cs.DB↗
arXiv2026

SUM: Unified Geometric Surgery on Spatio-Temporal Adaptation Vectors for Federated Class Incremental Learning

Real-world intelligent systems often require both distributed collaboration across data-isolated clients and continual adaptation to evolving tasks. This setting naturally gives rise to Federated Class Incremental Learning (FCIL), which combines Federated Learning (FL) and Continual Learning (CL). However, their combination introduces two coupled sources of interference: spatial interference from heterogeneous clients and temporal interference from sequential tasks, jointly leading to Spatial-Temporal Catastrophic Forgetting (ST-CF). Existing approaches typically address spatial and temporal interference with separate mechanisms, often incurring additional client-side computation or communication, while leaving directional interactions among updates during aggregation unregulated. In this paper, we reinterpret FCIL as a unified multi-task learning problem, where both client and task updates are represented as adaptation vectors in a shared parameter space. Based on this view, we propose Unified Geometric Surgery on Spatio-Temporal Adaptation Vectors (SUM), a purely server-side framework that performs geometric surgery on adaptation vectors during aggregation. Spatial SUM mitigates client-level interference within each round, while causal online temporal SUM removes cross-task interference over time without additional client-side computation, communication, or memory beyond standard federated training. Empirically, SUM achieves up to 22% improvement over prior FCIL methods across diverse vision and language benchmarks while remaining robust to unreliable clients and maintaining computational efficiency.

cs.LG↗
arXiv2026

Revisiting Filtered ANN Benchmarks: A Hardness-Controlled Benchmark Generator for Realistic Evaluation

Filtered approximate nearest neighbor (FANN) search must satisfy both vector similarity and structured predicates, yet evaluations remain brittle because real hybrid workloads are rarely shareable and existing benchmarks rely on ad-hoc synthetic or semi-real constructions. We argue that realism hinges on execution-driven query difficulty: failures in early filtering trigger over-fetching of additional candidates, shaping latency, throughput, and recall. Building on this insight, we propose $\alpha$-Hardness, a query-level hardness metric that models the conditional execution chain via the over-fetch factor and extends naturally to strategy-conditioned settings. Across diverse datasets and hybrid strategies, $\alpha$-Hardness exhibits strong monotonic alignment with empirical performance, while common proxies such as selectivity or attribute-vector correlation are frequently unstable or strategy-inconsistent. We further introduce HCBGen, a hardness-controlled benchmark generator that uses $\alpha$-Hardness as an explicit control signal to synthesize workloads under coarse bias modes or to match a target hardness profile. Our experiments show that widely used benchmarks occupy a narrow, relatively easy portion of the hardness spectrum, masking robustness gaps that emerge under harder queries. Finally, we demonstrate that matching hardness distributions enables privacy-preserving proxy workloads that closely reproduce performance trends, bridging research benchmarks and real evaluation.

cs.DB↗
arXiv2026

MI-CXR: A Benchmark for Longitudinal Reasoning over Multi-Interval Chest X-rays

Longitudinal chest X-ray (CXR) interpretation requires reasoning over disease evolution across multiple patient visits, yet most existing medical VQA benchmarks focus on single images or short-horizon image pairs. We introduce MI-CXR, a benchmark for standardized evaluation of Multi-Interval longitudinal reasoning over multi-visit CXR sequences, without requiring free-form report generation or additional clinical context. MI-CXR comprises five-way multiple-choice questions over five-visit patient timelines and instantiates three complementary task families: Temporal Event Localization, Interval-wise Change Reasoning, and Global Trajectory Summarization, which assess clinically grounded visual reasoning over time. Evaluating 14 state-of-the-art vision-language models (VLMs) shows low overall performance, with an average accuracy of 29.3%, only modestly above random guessing. Using stage-wise diagnostic probing, we find that models often produce locally plausible interval descriptions but fail to enforce temporal constraints or compose evidence into globally consistent decisions over the full timeline. These findings reveal key limitations of current VLMs and establish MI-CXR as a principled benchmark for longitudinal medical reasoning. The benchmark is available at https://github.com/AIDASLab/MI-CXR

cs.CV↗
arXiv2026

Report of the 5th PVUW Challenge: Towards More Diverse Modalities in Pixel-Level Understanding

This report summarizes the objectives, datasets, and top-performing methodologies of the 2026 Pixel-level Video Understanding in the Wild (PVUW) Challenge, hosted at CVPR 2026, which evaluates state-of-the-art models under highly unconstrained conditions. To provide a comprehensive assessment, the 2026 edition features three specialized tracks: the MOSE track for tracking objects within densely cluttered and severely occluded scenarios; the MeViS-Text track for localizing targets via motion-focused linguistic expressions; and the newly inaugurated MeViS-Audio track, which pioneers acoustic-driven object segmentation. By introducing previously unreleased challenging data and analyzing the cutting-edge, multimodal solutions submitted by participants, this report highlights the community's latest technical advancements and charts promising future directions for robust video scene comprehension.

cs.CV↗
arXiv2026

VIRST: Video-Instructed Reasoning Assistant for SpatioTemporal Segmentation

Referring Video Object Segmentation (RVOS) aims to segment target objects in videos based on natural language descriptions. However, fixed keyframe-based approaches that couple a vision language model with a separate propagation module often fail to capture rapidly changing spatiotemporal dynamics and to handle queries requiring multi-step reasoning, leading to sharp performance drops on motion-intensive and reasoning-oriented videos beyond static RVOS benchmarks. To address these limitations, we propose VIRST (Video-Instructed Reasoning Assistant for Spatio-Temporal Segmentation), an end-to-end framework that unifies global video reasoning and pixel-level mask prediction within a single model. VIRST bridges semantic and segmentation representations through the Spatio-Temporal Fusion (STF), which fuses segmentation-aware video features into the vision-language backbone, and employs the Temporal Dynamic Anchor Updater to maintain temporally adjacent anchor frames that provide stable temporal cues under large motion, occlusion, and reappearance. This unified design achieves state-of-the-art results across diverse RVOS benchmarks under realistic and challenging conditions, demonstrating strong generalization to both referring and reasoning oriented settings. The code and checkpoints are available at https://github.com/AIDASLab/VIRST.

cs.CV↗
arXiv2026

MEDIC-AD: Towards Medical Vision-Language Model's Clinical Intelligence

Lesion detection, symptom tracking, and visual explainability are central to real-world medical image analysis, yet current medical Vision-Language Models (VLMs) still lack mechanisms that translate their broad knowledge into clinically actionable outputs. To bridge this gap, we present MEDIC-AD, a clinically oriented VLM that strengthens these three capabilities through a stage-wise framework. First, learnable anomaly-aware tokens ( ) encourage the model to focus on abnormal regions and build more discriminative lesion centered representations. Second, inter image difference tokens ( ) explicitly encode temporal changes between studies, allowing the model to distinguish worsening, improvement, and stability in disease burden. Finally, a dedicated explainability stage trains the model to generate heatmaps that highlight lesion-related regions, offering clear visual evidence that is consistent with the model's reasoning. Through our staged design, MEDIC-AD steadily boosts performance across anomaly detection, symptom tracking, and anomaly segmentation, achieving state-of-the-art results compared with both closed source and medical-specialized baselines. Evaluations on real longitudinal clinical data collected from real hospital workflows further show that MEDIC-AD delivers stable predictions and clinically faithful explanations in practical patient-monitoring and decision-support workflows

cs.CV↗
arXiv2026

3rd Place of MeViS-Audio Track of the 5th PVUW: VIRST-Audio

Audio-based Referring Video Object Segmentation (ARVOS) requires grounding audio queries into pixel-level object masks over time, posing challenges in bridging acoustic signals with spatio-temporal visual representations. In this report, we present VIRST-Audio, a practical framework built upon a pretrained RVOS model integrated with a vision-language architecture. Instead of relying on audio-specific training, we convert input audio into text using an ASR module and perform segmentation using text-based supervision, enabling effective transfer from text-based reasoning to audio-driven scenarios. To improve robustness, we further incorporate an existence-aware gating mechanism that estimates whether the referred target object is present in the video and suppresses predictions when it is absent, reducing hallucinated masks and stabilizing segmentation behavior. We evaluate our approach on the MeViS-Audio track of the 5th PVUW Challenge, where VIRST-Audio achieves 3rd place, demonstrating strong generalization and reliable performance in audio-based referring video segmentation.

cs.CV↗
arXiv2026

Dynin-Omni: Omnimodal Unified Large Diffusion Language Model

We present Dynin-Omni, the first masked-diffusion-based omnimodal foundation model that unifies text, image, and speech understanding and generation, together with video understanding, within a single architecture. Unlike autoregressive unified models that serialize heterogeneous modalities, or compositional unified models that require orchestration with external modality-specific decoders, Dynin-Omni natively formulates omnimodal modeling as masked diffusion over a shared discrete token space, enabling iterative refinement under bidirectional context. Dynin-Omni adopts a multi-stage training strategy with model-merging-based modality expansion and omnimodal alignment. We evaluate Dynin-Omni across 19 multimodal benchmarks spanning language reasoning, image generation and editing, video understanding, and speech recognition and synthesis. Dynin-Omni achieves 87.6 on GSM8K, 1733.6 on MME-P, 61.4 on VideoMME, 0.87 on GenEval, and 2.1 WER on LibriSpeech test-clean, consistently outperforming existing open-source unified models while remaining competitive with strong modality-specific expert systems. These results demonstrate the potential of masked diffusion as a unified paradigm for any-to-any modeling, providing a flexible foundation for real-time omnimodal systems, unified cross-modal retrieval and generation, and embodied multimodal agents.

cs.CL↗
arXiv2026

RFEval: Benchmarking Reasoning Faithfulness under Counterfactual Reasoning Intervention in Large Reasoning Models

Large Reasoning Models (LRMs) exhibit strong performance, yet often produce rationales that sound plausible but fail to reflect their true decision process, undermining reliability and trust. We introduce a formal framework for reasoning faithfulness, defined by two testable conditions: stance consistency (a coherent stance linking reasoning to answer) and causal influence (the stated reasoning causally drives the answer under output-level interventions), explicitly decoupled from accuracy. To operationalize this, we present RFEval, a benchmark of 7,186 instances across seven tasks that probes faithfulness via controlled, output-level counterfactual interventions. Evaluating twelve open-source LRMs, we find unfaithfulness in 49.7% of outputs, predominantly from stance inconsistency. Failures are concentrated in brittle, convergent domains such as math and code, and correlate more with post-training regimes than with scale: within-family ablations indicate that adding current RL-style objectives on top of supervised fine-tuning can reduce reasoning faithfulness, even when accuracy is maintained. Crucially, accuracy is neither a sufficient nor a reliable proxy for faithfulness: once controlling for model and task, the accuracy-faithfulness link is weak and statistically insignificant. Our work establishes a rigorous methodology for auditing LRM reliability and shows that trustworthy AI requires optimizing not only for correct outcomes but also for the structural integrity of the reasoning process. Our code and dataset can be found at project page: https://aidaslab.github.io/RFEval/

cs.AI↗
arXiv2026

MATA: Multi-Agent Framework for Reliable and Flexible Table Question Answering

Recent advances in Large Language Models (LLMs) have significantly improved table understanding tasks such as Table Question Answering (TableQA), yet challenges remain in ensuring reliability, scalability, and efficiency, especially in resource-constrained or privacy-sensitive environments. In this paper, we introduce MATA, a multi-agent TableQA framework that leverages multiple complementary reasoning paths and a set of tools built with small language models. MATA generates candidate answers through diverse reasoning styles for a given table and question, then refines or selects the optimal answer with the help of these tools. Furthermore, it incorporates an algorithm designed to minimize expensive LLM agent calls, enhancing overall efficiency. MATA maintains strong performance with small, open-source models and adapts easily across various LLM types. Extensive experiments on two benchmarks of varying difficulty with ten different LLMs demonstrate that MATA achieves state-of-the-art accuracy and highly efficient reasoning while avoiding excessive LLM inference. Our results highlight that careful orchestration of multiple reasoning pathways yields scalable and reliable TableQA. The code is available at https://github.com/AIDASLab/MATA.

cs.CL↗
arXiv2026

Is Retraining-Free Enough? The Necessity of Router Calibration for Efficient MoE Compression

Mixture-of-Experts (MoE) models scale capacity efficiently, but their massive parameter footprint creates a deployment-time memory bottleneck. We organize retraining-free MoE compression into three paradigms - Expert Pruning, Expert Editing, and Expert Merging - and show that persistent post-compression degradation largely stems from a neglected factor: router-expert mismatch when experts are changed but the router is left untouched. We argue that effective retraining-free compression should avoid updating expert parameters while allowing lightweight router calibration. To this end, we propose Router Knowledge Distillation (Router KD), which updates only a tiny fraction of parameters (the router) by distilling the original model's next-token distribution on unlabeled calibration data. Experiments across representative methods in all three paradigms demonstrate consistent performance recovery, with substantially larger gains in fine-grained MoEs (many small experts) than in coarse-grained MoEs due to their more complex routing decision boundaries.

cs.LG↗
arXiv2026

VALUEFLOW: Toward Pluralistic and Steerable Value-based Alignment in Large Language Models

Aligning Large Language Models (LLMs) with the diverse spectrum of human values remains a central challenge: preference-based methods often fail to capture deeper motivational principles. Value-based approaches offer a more principled path, yet three gaps persist: extraction often ignores hierarchical structure, evaluation detects presence but not calibrated intensity, and the steerability of LLMs at controlled intensities remains insufficiently understood. To address these limitations, we introduce VALUEFLOW, the first unified framework that spans extraction, evaluation, and steering with calibrated intensity control. The framework integrates three components: (i) HIVES, a hierarchical value embedding space that captures intra- and cross-theory value structure; (ii) the Value Intensity DataBase (VIDB), a large-scale resource of value-labeled texts with intensity estimates derived from ranking-based aggregation; and (iii) an anchor-based evaluator that produces consistent intensity scores for model outputs by ranking them against VIDB panels. Using VALUEFLOW, we conduct a comprehensive large-scale study across ten models and four value theories, identifying asymmetries in steerability and composition laws for multi-value control. This paper establishes a scalable infrastructure for evaluating and controlling value intensity, advancing pluralistic alignment of LLMs.

cs.AI↗
arXiv2025

MatKV: Trading Compute for Flash Storage in LLM Inference

We observe two major trends in LLM-based generative AI: (1) inference is becoming the dominant factor in terms of cost and power consumption, surpassing training, and (2) retrieval augmented generation (RAG) is becoming prevalent. When processing long inputs in RAG, the prefill phase of computing the key-value vectors of input text is energy-intensive and time-consuming even with high-end GPUs. Thus, it is crucial to make the prefill phase in RAG inference efficient. To address this issue, we propose MatKV, a scheme that precomputes the key-value vectors (KVs) of RAG objects (e.g., documents), materializes them in inexpensive but fast and power-efficient flash storage, and reuses them at inference time instead of recomputing the KVs using costly and power-inefficient GPU. Experimental results using Hugging Face's Transformers library across state-of-the-art GPUs and flash memory SSDs confirm that, compared to full KV computation on GPUs, MatKV reduces both inference time and power consumption by half for RAG workloads, without severely impacting accuracy in the question-answering task. Furthermore, we demonstrate that MatKV enables additional optimizations in two ways. First, a GPU can decode text while simultaneously loading the materialized KVs for the next instance, reducing load latency. Second, since decoding speed is less sensitive to GPU performance than KV computation, low-end GPUs can be leveraged for decoding without significantly compromising speed once the materialized KVs are loaded into GPU memory. These findings underscore MatKV's potential to make large-scale generative AI applications more cost-effective, power-efficient, and accessible across a wider range of tasks and hardware environments.

cs.DC↗
arXiv2025

Don't Let It Fade: Preserving Edits in Diffusion Language Models via Token Timestep Allocation

While diffusion language models (DLMs) enable fine-grained refinement, their practical controllability remains fragile. We identify and formally characterize a central failure mode called update forgetting, in which uniform and context agnostic updates induce token level fluctuations across timesteps, erasing earlier semantic edits and disrupting the cumulative refinement process, thereby degrading fluency and coherence. As this failure originates in uniform and context agnostic updates, effective control demands explicit token ordering. We propose Token Timestep Allocation (TTA), which realizes soft and semantic token ordering via per token timestep schedules: critical tokens are frozen early, while uncertain tokens receive continued refinement. This timestep based ordering can be instantiated as either a fixed policy or an adaptive policy driven by task signals, thereby supporting a broad spectrum of refinement strategies. Because it operates purely at inference time, it applies uniformly across various DLMs and naturally extends to diverse supervision sources. Empirically, TTA improves controllability and fluency: on sentiment control, it yields more than 20 percent higher accuracy and nearly halves perplexity using less than one fifth the steps; in detoxification, it lowers maximum toxicity (12.2 versus 14.5) and perplexity (26.0 versus 32.0). Together, these results demonstrate that softened ordering via timestep allocation is the critical lever for mitigating update forgetting and achieving stable and controllable diffusion text generation.

cs.CL↗
arXiv2025

Exploring and Leveraging Class Vectors for Classifier Editing

Image classifiers play a critical role in detecting diseases in medical imaging and identifying anomalies in manufacturing processes. However, their predefined behaviors after extensive training make post hoc model editing difficult, especially when it comes to forgetting specific classes or adapting to distribution shifts. Existing classifier editing methods either focus narrowly on correcting errors or incur extensive retraining costs, creating a bottleneck for flexible editing. Moreover, such editing has seen limited investigation in image classification. To overcome these challenges, we introduce Class Vectors, which capture class-specific representation adjustments during fine-tuning. Whereas task vectors encode task-level changes in weight space, Class Vectors disentangle each class's adaptation in the latent space. We show that Class Vectors capture each class's semantic shift and that classifier editing can be achieved either by steering latent features along these vectors or by mapping them into weight space to update the decision boundaries. We also demonstrate that the inherent linearity and orthogonality of Class Vectors support efficient, flexible, and high-level concept editing via simple class arithmetic. Finally, we validate their utility in applications such as unlearning, environmental adaptation, adversarial defense, and adversarial trigger optimization.

cs.CV↗