Searcharxiv⌕ Search

arXiv subjects

Minji Kim

Publications and source records attributed to Minji Kim.

At least 19 recordsLinked to original sources
arXiv2026

Reward Inflation: A Healthy Stimulus for Reinforcement Learning

Reward serves as the primary learning signal in reinforcement learning (RL). However, while reward magnitudes are typically held fixed throughout training, their temporal modulation remains underexplored. In this paper, we propose reward inflation, a gradual scaling of rewards over the course of training, and show that it can act as a healthy stimulus for RL. Theoretically, reward inflation induces an implicit recency weighting that upweights recent transitions during policy updates, enabling faster adaptation. We further show that, by sustaining gradient signals as the policy saturates, reward inflation suppresses the emergence of dormant neurons and helps preserve plasticity. Empirical results on ALE games and MuJoCo tasks corroborate these findings, showing that an appropriate level of reward inflation benefits a broad range of tasks. Finally, we introduce Fed, an adaptive variant that adjusts the inflation level on the fly, and find that it often improves upon fixed inflation.

cs.LG↗
arXiv2026

Bridging LLM Serving and CXL-SSDs with Chunk-Aware KV Cache Management

NAND-backed storage offers the capacity needed to scale LLM prefix caching, but its block I/O path incurs CPU cache contention and host-DRAM staging in addition to NAND latency. Our characterization shows that these interface costs persist even with DRAM as the storage medium, motivating CXL-SSDs for byte-addressable access to NAND-backed capacity. Surprisingly, however, a stock CXL-SSD remains about 3$\times$ slower than local DRAM and no faster than an NVMe SSD, while generic prefetching provides little benefit. We present LM-CXD, a CXL-SSD specialized for LLM prefix caching. LM-CXD bridges the semantic gap between the serving engine, which knows which KV chunks will be consumed, and the device, which controls their placement and movement. It makes KV chunks device-visible I/O units, exposes NAND-to-DRAM progress to the serving engine, and uses device DRAM as a GPU-accessible buffer. LM-CXD further coordinates request scheduling with windowed prefetching and pipelines layerwise KV movement with GPU computation to hide NAND latency under limited device DRAM. Across five LLM models, LM-CXD reduces average TTFT over a stock CXL-SSD by up to 2.6$\times$ with compute asynchronous prefetching and 4.03$\times$ with layerwise prefetching, achieving TTFT within 1.5$\times$ of local DRAM on average.

cs.AR↗
arXiv2026

GALoc: Gravity Aligned Wireframes for Depth-Free Monocular Floorplan Localization

Floorplans are compact, appearance-invariant maps ideal for indoor localization, yet existing methods rely on depth networks that are brittle in cluttered scenes. We propose GALoc, a geometry-first framework that replaces depth prediction with gravity-aligned wireframes that satisfy verticality and coplanarity by construction. Given monocular RGB, camera intrinsics, relative poses, and IMU orientation, GALoc constructs a linear constraint matrix encoding verticality and coplanarity, and finds the camera gauge minimizing its smallest singular value via global search. The rectified wireframes are projected into bird's-eye-view layouts through a closed-form, FOV-consistent transformation and matched against the floorplan via metric-free SE(2) search. We evaluate end-to-end on Structured3D, with calibrated noise on Gibson, and on real-world author-collected sequences. When sufficient wall geometry is visible, GALoc matches or outperforms depth-based baselines -- achieving 88% sequential localization success at 0.1m over 100-step sequences on Gibson vs the baseline's 68% -- while abstaining in structure-blind scenes.

cs.CV↗
arXiv2026

Refuse without Refusal: A Structural Analysis of Safety-Tuning Responses for Reducing False Refusals in Language Models

Striking a balance between helpfulness and safety remains a fundamental challenge in aligning large language models. To achieve this balance, models should refuse harmful queries (e.g., "How do I shoot someone?") while remaining responsive to benign inputs, even those superficially resembling harmful queries (e.g., "Where can I shoot a good photo?"). However, models often struggle to distinguish genuinely harmful queries from benign queries that contain superficially risky language, resulting in false refusals. In this paper, we address the issue by decomposing a response in the safety-tuning dataset into two distinct components: (i) a boilerplate refusal statement and (ii) a rationale explaining the refusal. Our experiments and analyses show that refusal statements impede accurate discrimination between harmful and benign queries by inducing reliance on superficial cues. In contrast, training solely on rationales reduces false refusals while maintaining a comparable level of safety performance. Rationale-Only benefits also appear in our ICL configuration and remain compatible with the evaluated inference-time mitigation methods. The results emphasize the necessity of precisely curated, fine-grained safety supervision datasets and outline directions for constructing aligned agents that better reconcile helpfulness with safety.

cs.CL↗
arXiv2026

Knowing What Not to Answer: Selective Non-Compliance in Vision-Language Models

Vision-language models (VLMs) are expected to respond helpfully to appropriate requests while withholding compliance with requests that are incorrect, unsafe, infeasible, or unanswerable. However, existing benchmarks predominantly evaluate non-compliance at the level of the query as a whole, assuming that each request either warrants compliance or requires withholding compliance. In practice, real-world queries can contain a mixture of answerable content and components for which compliance should be withheld. In this paper, we introduce KoNA, a benchmark for evaluating selective non-compliance in VLMs across five categories: False Premise, Visual Inaccessibility, Universal Unknown, Task Feasibility, and Safety. Each task evaluates two capabilities: query-level non-compliance and component-level non-compliance under paired single and compound queries. Our evaluation across diverse VLMs shows that models often fail to refuse, correct, or abstain appropriately, and these failures become more pronounced when queries require selective non-compliance. To address this challenge, we fine-tune VLMs using KoNA examples that require selective non-compliance, together with a fully answerable set that should receive direct answers. Our fine-tuned models achieve substantial improvements in non-compliance accuracy while largely maintaining performance on fully answerable tasks. These results suggest that the fine-tuned models can distinguish between answerable components and those requiring non-compliance and respond in a task-appropriate manner.

cs.CL↗
arXiv2026

Benchmarking Vision-Language Models for Automated Pathology Diagnosis and Report Generation

The rapid advancement of vision-language models (VLMs) has accelerated progress in computational pathology; however, whole-slide image (WSI)-based pathology report generation remains limited by the scarcity of large-scale WSI--report datasets and the complexity of mapping spatially distributed visual patterns to structured clinical text. To address this, we introduce a clinically curated Pan-Asia WSI--report dataset of approximately 10,500 pairs from five institutions and establish the REG 2025 benchmark through a MICCAI challenge for systematic evaluation of multimodal models. We analyze submitted methods spanning pretrained VLMs, multiple-instance learning frameworks, hierarchical expert models, retrieval-augmented generation, and cross-modal Transformers. Rather than indicating that VLM use alone was sufficient for superior performance, the results suggest that top-performing methods benefited from structured report representations, hierarchical diagnostic decomposition, and effective multimodal grounding. We identify key limitations, including instability in quantitative attribute estimation (e.g., numeric hallucination) and a tendency toward diagnostic overspecification, with some errors resembling known diagnostic pitfalls in routine pathology. These findings establish REG 2025 as a benchmark for evaluating WSI-based structured report generation and vision-language understanding in computational pathology, providing insights for the design of clinically grounded multimodal pathology models.

cs.CV↗
arXiv2026

Hyper-reduction methods for accelerating nonlinear finite element simulations: open source implementation and reproducible benchmarks

Hyper-reduction methods have gained increasing attention for their potential to accelerate reduced order models for nonlinear systems, yet their comparative accuracy and computational efficiency are not well understood. Motivated by this gap, we evaluate a range of hyper-reduction techniques for nonlinear finite element models across benchmark problems of varying complexity, assessing the inevitable tradeoff between accuracy and speedup. More specifically, we consider interpolation methods based on the gappy proper orthogonal decomposition as well as the empirical quadrature procedure (EQP), and apply them to the hyper-reduction of problems in nonlinear diffusion, nonlinear elasticity and Lagrangian hydrodynamics. Our numerical results are generated using the open source libROM, Laghos and MFEM numerical libraries. Our findings reveal that the comparative performance between hyper-reduction methods depends on both the problem and the choice of time integration method. The EQP method generally achieves lower relative errors than interpolation methods and is more efficient in terms of quadrature point usage, resulting in a lower wall time for the nonlinear diffusion and elasticity problems. However, its online computational cost is observed to be relatively high for Lagrangian hydrodynamics problems. Conversely, interpolation methods exhibit greater variability, especially with respect to the use of different time integration methods in the Lagrangian hydrodynamics problems. The presented results underscore the need for problem specific method selection to balance accuracy and efficiency, while also offering useful guidance for future comparisons and refinements of hyper-reduction techniques.

cs.MS↗
arXiv2026

REFINE-DP: Diffusion Policy Fine-tuning for Humanoid Loco-manipulation via Reinforcement Learning

Humanoid loco-manipulation requires coordinated task-space motion planning with stable loco-manipulation command tracking under complex robot-environment dynamics and long-horizon tasks. While diffusion policies (DPs) show promise for learning from demonstrations, deploying them on humanoids poses critical challenges: the motion planner trained offline is decoupled from the loco-manipulation controller, leading to poor command tracking, compounding distribution shift, and task failures. The common approach of scaling demonstration data is prohibitively expensive for high-dimensional humanoid systems. To address this challenge, we present REFINE-DP (REinforcement learning FINE-tuning of Diffusion Policy), a hierarchical framework that jointly optimizes a DP motion planner and an RL-based loco-manipulation controller. The DP is fine-tuned via a PPO-based diffusion policy gradient to improve task success rate, while the controller is simultaneously updated to accurately track the planner's evolving command distribution, reducing the distributional mismatch that degrades motion quality. We validate REFINE-DP on a humanoid robot performing loco-manipulation tasks, including door traversal and long-horizon object transport. REFINE-DP achieves an over 90% success rate in simulation, even in out-of-distribution cases not seen in the pre-training data, and enables real-world execution without privileged state information. Our proposed method substantially outperforms pre-trained DP baselines and demonstrates that RL fine-tuning is key to reliable humanoid loco-manipulation. https://refine-dp.github.io/REFINE-DP/

cs.RO↗
arXiv2026

Wake up for Touch! Mask-isolated Tactile Alignment Learning in MLLMs

Touch supplies the physical grounding needed to perceive intrinsic material properties, such as friction and compliance, that vision alone often cannot resolve. Recent efforts for equipping multimodal LLMs with this tactile sense, however, expose a zero-sum trade-off: the limited parameter budget of compact models forces a choice between acquiring the new sensory modality and preserving the established vision-language reasoning. We present Splash, a mask-isolated tactile alignment learning framework for MLLMs. Splash quantifies the significance of each pretrained parameter, and partitions the parameter space into a dormant and critical subspace. While the frozen critical subspace acts as a stable anchor to safeguard general visual knowledge, Splash updates the isolated dormant subspace to internalize tactile alignment towards LLMs. This selective, non-destructive expansion effectively prevents catastrophic forgetting and ensures non-destructive modality expansion. Extensive experiments show that Splash effectively achieves tactile reasoning without additional inference overhead in the LLM part, demonstrating state-of-the-art performance on visuo-tactile benchmarks, including SSVTP, TVL, and TacQuad, while preserving its original general-purpose capabilities.

cs.CV↗
arXiv2026

Rule-based autocorrection of Piping and Instrumentation Diagrams (P&IDs) on graphs

A piping and instrumentation diagram (P&ID) is a central reference document in chemical process engineering. Currently, chemical engineers manually review P&IDs through visual inspection to find and rectify errors. However, engineering projects can involve hundreds to thousands of P&ID pages, creating a significant revision workload. This study proposes a rule-based method to support engineers with error detection and correction in P&IDs. The method is based on a graph representation of P&IDs, enabling automated error detection and correction, i.e., autocorrection, through rule graphs. We use our pyDEXPI Python package to generate P&ID graphs from DEXPI-standard P&IDs. In this study, we developed 33 rules based on chemical engineering knowledge and heuristics, with five selected rules demonstrated as examples. A case study on an illustrative P&ID validates the reliability and effectiveness of the rule-based autocorrection method in revising P&IDs.

cs.CE↗
arXiv2026

LightSplat: Fast and Memory-Efficient Open-Vocabulary 3D Scene Understanding in Five Seconds

Open-vocabulary 3D scene understanding enables users to segment novel objects in complex 3D environments through natural language. However, existing approaches remain slow, memory-intensive, and overly complex due to iterative optimization and dense per-Gaussian feature assignments. To address this, we propose LightSplat, a fast and memory-efficient training-free framework that injects compact 2-byte semantic indices into 3D representations from multi-view images. By assigning semantic indices only to salient regions and managing them with a lightweight index-feature mapping, LightSplat eliminates costly feature optimization and storage overhead. We further ensure semantic consistency and efficient inference via single-step clustering that links geometrically and semantically related masks in 3D. We evaluate our method on LERF-OVS, ScanNet, and DL3DV-OVS across complex indoor-outdoor scenes. As a result, LightSplat achieves state-of-the-art performance with up to 50-400x speedup and 64x lower memory, enabling scalable language-driven 3D understanding. For more details, visit our project page https://vision3d-lab.github.io/lightsplat/.

cs.CV↗
arXiv2026

Parametric multi-fidelity Monte Carlo estimation with applications to extremes

In a multi-fidelity setting, data are available from two sources, high- and low-fidelity. Low-fidelity data has larger size and can be leveraged to make more efficient inference about quantities of interest, e.g. the mean, for high-fidelity variables. In this work, such multi-fidelity setting is studied when the goal is to fit more efficiently a parametric model to high-fidelity data. Three multi-fidelity parameter estimation methods are considered, joint maximum likelihood, (multi-fidelity) moment estimation and (multi-fidelity) marginal maximum likelihood, and are illustrated on several parametric models, with the focus on parametric families used in extreme value analysis. An application is also provided concerning quantification of occurrences of extreme ship motions generated by two computer codes of varying fidelity.

stat.ME↗
arXiv2026

Map the Flow: Revealing Hidden Pathways of Information in VideoLLMs

Video Large Language Models (VideoLLMs) extend the capabilities of vision-language models to spatiotemporal inputs, enabling tasks such as video question answering (VideoQA). Despite recent advances in VideoLLMs, their internal mechanisms on where and how they extract and propagate video and textual information remain less explored. In this study, we investigate the internal information flow of VideoLLMs using mechanistic interpretability techniques. Our analysis reveals consistent patterns across diverse VideoQA tasks: (1) temporal reasoning in VideoLLMs initiates with active cross-frame interactions in early-to-middle layers, (2) followed by progressive video-language integration in middle layers. This is facilitated by alignment between video representations and linguistic embeddings containing temporal concepts. (3) Upon completion of this integration, the model is ready to generate correct answers in middle-to-late layers. (4) Based on our analysis, we show that VideoLLMs retain their VideoQA performance by selecting these effective information pathways while suppressing a substantial amount of attention edges, e.g., 58% in LLaVA-NeXT-7B-Video-FT. These findings provide a blueprint for how VideoLLMs perform temporal reasoning and offer practical insights for improving model interpretability and downstream generalization. Our project page with the source code is available at https://map-the-flow.github.io

cs.CV↗
arXiv2026

ReVision: A Dataset and Baseline VLM for Privacy-Preserving Task-Oriented Visual Instruction Rewriting

Efficient and privacy-preserving multimodal interaction is essential as AR, VR, and modern smartphones with powerful cameras become primary interfaces for human-computer communication. Existing powerful large vision-language models (VLMs) enabling multimodal interaction often rely on cloud-based processing, raising significant concerns about (1) visual privacy by transmitting sensitive vision data to servers, and (2) their limited real-time, on-device usability. This paper explores Visual Instruction Rewriting, a novel approach that transforms multimodal instructions into text-only commands, allowing seamless integration of lightweight on-device instruction rewriter VLMs (250M parameters) with existing conversational AI systems, enhancing vision data privacy. To achieve this, we present a dataset of over 39,000 examples across 14 domains and develop a compact VLM, pretrained on image captioning datasets and fine-tuned for instruction rewriting. Experimental results, evaluated through NLG metrics such as BLEU, METEOR, and ROUGE, along with semantic parsing analysis, demonstrate that even a quantized version of the model (<500MB storage footprint) can achieve effective instruction rewriting, thus enabling privacy-focused, multimodal AI applications.

cs.CL↗
arXiv2025

BRIC: Bridging Kinematic Plans and Physical Control at Test Time

We propose BRIC, a novel test-time adaptation (TTA) framework that enables long-term human motion generation by resolving execution discrepancies between diffusion-based kinematic motion planners and reinforcement learning-based physics controllers. While diffusion models can generate diverse and expressive motions conditioned on text and scene context, they often produce physically implausible outputs, leading to execution drift during simulation. To address this, BRIC dynamically adapts the physics controller to noisy motion plans at test time, while preserving pre-trained skills via a loss function that mitigates catastrophic forgetting. In addition, BRIC introduces a lightweight test-time guidance mechanism that steers the diffusion model in the signal space without updating its parameters. By combining both adaptation strategies, BRIC ensures consistent and physically plausible long-term executions across diverse environments in an effective and efficient manner. We validate the effectiveness of BRIC on a variety of long-term tasks, including motion composition, obstacle avoidance, and human-scene interaction, achieving state-of-the-art performance across all tasks.

cs.CV↗
arXiv2025

Explorable Ideas: Externalizing Ideas as Explorable Environments

Working with abstract information often relies on static, symbolic representations that constrain exploration. We introduce Explorable Ideas, a framework that externalizes abstract concepts into explorable environments where physical navigation coordinates conceptual exploration. To investigate its practical value, we designed Idea Islands, a VR probe for ideation tasks, and conducted two controlled studies with 19 participants. Results show that overview perspectives foster strategic breadth while immersion sustains engagement through embodied presence, and that seamless transitions enable flexible workflows combining both modes. These findings validate the framework's design considerations and yield design implications for building future systems that treat information as explorable territory across creative, educational, and knowledge-intensive domains.

cs.HC↗
arXiv2025

Revealing the Inherent Instructability of Pre-Trained Language Models

Instruction tuning -- supervised fine-tuning using instruction-response pairs -- is a key step in making pre-trained large language models (LLMs) instructable. Meanwhile, LLMs perform multitask learning during their pre-training, acquiring extensive knowledge and capabilities. We hypothesize that the pre-training stage can enable them to develop the ability to comprehend and address instructions. To verify this, we propose Response Tuning (RT), which removes the instruction and its corresponding mapping to the response from instruction tuning. Instead, it focuses solely on establishing a response distribution. Our experiments demonstrate that RT models, trained only on responses, can effectively respond to a wide range of instructions akin to their instruction-tuned counterparts. In addition, we observe that the models can recognize and reject unsafe queries after learning a safety policy only from the response data. Furthermore, we find that these observations extend to an in-context learning setting. These findings support our hypothesis, highlighting the extensive inherent capabilities of pre-trained LLMs.

cs.CL↗
arXiv2025

Enabling Chatbots with Eyes and Ears: An Immersive Multimodal Conversation System for Dynamic Interactions

As chatbots continue to evolve toward human-like, real-world, interactions, multimodality remains an active area of research and exploration. So far, efforts to integrate multimodality into chatbots have primarily focused on image-centric tasks, such as visual dialogue and image-based instructions, placing emphasis on the "eyes" of human perception while neglecting the "ears", namely auditory aspects. Moreover, these studies often center around static interactions that focus on discussing the modality rather than naturally incorporating it into the conversation, which limits the richness of simultaneous, dynamic engagement. Furthermore, while multimodality has been explored in multi-party and multi-session conversations, task-specific constraints have hindered its seamless integration into dynamic, natural conversations. To address these challenges, this study aims to equip chatbots with "eyes and ears" capable of more immersive interactions with humans. As part of this effort, we introduce a new multimodal conversation dataset, Multimodal Multi-Session Multi-Party Conversation ($M^3C$), and propose a novel multimodal conversation model featuring multimodal memory retrieval. Our model, trained on the $M^3C$, demonstrates the ability to seamlessly engage in long-term conversations with multiple speakers in complex, real-world-like settings, effectively processing visual and auditory inputs to understand and respond appropriately. Human evaluations highlight the model's strong performance in maintaining coherent and dynamic interactions, demonstrating its potential for advanced multimodal conversational agents.

cs.CL↗