Beomjun Kim

Woomin Song, Beomjun Kim, Daewon Choi et al.

A common failure mode in long-horizon agentic test-time scaling is error propagation, where factual errors or invalid deductions introduced at intermediate steps persist in the agent's belief state and contaminate later reasoning. Existing test-time scaling methods provide limited control over this process, as they often rely on agents to detect their own mistakes, select among flawed trajectories, or refine solutions only after errors have already shaped the reasoning path. We propose ExComm, a communication protocol for exploration-stage agentic test-time scaling. ExComm is motivated by the empirical observation that the majority of intermediate errors in parallel agentic reasoning produce detectable cross-agent factual conflicts. Leveraging the iterative structure of agentic workflows, ExComm periodically audits agent belief states to detect such conflicts, resolves them through a dedicated tool-based verification loop, and returns concise, targeted feedback to the involved agents. Corrections are incorporated through soft belief updates, which append verified feedback rather than overwriting existing beliefs. Furthermore, to prevent collapsing trajectory diversity due to communication, ExComm further introduces a trajectory diversification module that redirects redundant trajectories toward orthogonal strategies. Experiments on AIME 2024, AIME 2025, and GAIA with Gemini-2.5-Flash-Lite and Qwen3.5-4B show that ExComm consistently outperforms strong test-time scaling baselines, achieving average performance gains of 5.7% and 5.0% over the best-performing baselines, respectively. Further analyses demonstrate improved error recovery, favorable scaling behavior, stronger diversity than adapted communication baselines, and the best performance-cost trade-off among the evaluated methods.

Jeonghoon Kim, Byeongchan Lee, Cheonbok Park et al.

Selecting a layer normalization (LN) strategy that stabilizes training and speeds convergence in Transformers remains difficult, even for today's large language models (LLM). We present a comprehensive analytical foundation for understanding how different LN strategies influence training dynamics in large-scale Transformers. Until recently, Pre-LN and Post-LN have long dominated practices despite their limitations in large-scale training. However, several open-source models have recently begun silently adopting a third strategy without much explanation. This strategy places normalization layer peripherally around sublayers, a design we term Peri-LN. While Peri-LN has demonstrated promising performance, its precise mechanisms and benefits remain almost unexplored. Our in-depth analysis delineates the distinct behaviors of LN strategies, showing how each placement shapes activation variance and gradient propagation. To validate our theoretical insight, we conduct extensive experiments on Transformers up to $3.2$B parameters, showing that Peri-LN consistently achieves more balanced variance growth, steadier gradient flow, and convergence stability. Our results suggest that Peri-LN warrants broader consideration for large-scale Transformer architectures, providing renewed insights into the optimal placement of LN.

Hyunseok Lee, Jeonghoon Kim, Beomjun Kim et al.

Recent advances in Multimodal Large Language Models (MLLMs) have enabled autonomous agents to interact with computers via Graphical User Interfaces (GUIs), where accurately localizing the coordinates of interface elements (e.g., buttons) is often required for fine-grained actions. However, this remains significantly challenging, leading prior works to rely on large-scale web datasets to improve the grounding accuracy. In this work, we propose Reasoning Graphical User Interface Grounding for Data Efficiency (ReGUIDE), a novel and effective framework for web grounding that enables MLLMs to learn data efficiently through self-generated reasoning and spatial-aware criticism. More specifically, ReGUIDE learns to (i) self-generate a language reasoning process for the localization via online reinforcement learning, and (ii) criticize the prediction using spatial priors that enforce equivariance under input transformations. At inference time, ReGUIDE further boosts performance through a test-time scaling strategy, which combines spatial search with coordinate aggregation. Our experiments demonstrate that ReGUIDE significantly advances web grounding performance across multiple benchmarks, outperforming baselines with substantially fewer training data points (e.g., only 0.2% samples compared to the best open-sourced baselines).

Dongyoung Kim, Huiwon Jang, Myungkyu Koo et al.

While Vision-Language-Action models (VLAs) have shown remarkable progress toward human-like generalist robotic policies through the versatile intelligence (i.e. broad scene understanding and language-conditioned generalization) inherited from pre-trained Vision-Language Models, they still struggle with complex real-world tasks requiring broader functional capabilities (e.g. motion awareness, memory-aware decision making, and physical sensing). To address this, we introduce RLDX-1, a general-purpose robotic policy for dexterous manipulation built on the Multi-Stream Action Transformer (MSAT), an architecture that unifies these capabilities by integrating heterogeneous modalities through modality-specific streams with cross-modal joint self-attention. RLDX-1 further combines this architecture with system-level design choices, including synthesizing training data for rare manipulation scenarios, learning procedures specialized for human-like manipulation, and inference optimizations for real-time deployment. Through empirical evaluation, we show that RLDX-1 consistently outperforms recent frontier VLAs (e.g. $π_{0.5}$ and GR00T N1.6) across both simulation benchmarks and real-world tasks that require broad functional capabilities beyond general versatility. In particular, RLDX-1 shows superiority in ALLEX humanoid tasks by achieving success rates of 86.8% while $π_{0.5}$ and GR00T N1.6 achieve around 40%, highlighting the ability of RLDX-1 to control a high-DoF humanoid robot under diverse functional demands. Together, these results position RLDX-1 as a promising step toward reliable VLAs for complex, contact-rich, and dynamic real-world dexterous manipulation.

Beomjun Kim, Suhan Woo, Sejong Heo et al.

Bird's-Eye-View (BEV) map segmentation is one of the most important and challenging tasks in autonomous driving. Camera-only approaches have drawn attention as cost-effective alternatives to LiDAR, but they still fall behind LiDAR-Camera (LC) fusion-based methods. Knowledge Distillation (KD) has been explored to narrow this gap, but existing methods mainly enlarge the student model by mimicking the teacher's architecture, leading to higher inference cost. To address this issue, we introduce BridgeTA, a cost-effective distillation framework to bridge the representation gap between LC fusion and Camera-only models through a Teacher Assistant (TA) network while keeping the student's architecture and inference cost unchanged. A lightweight TA network combines the BEV representations of the teacher and student, creating a shared latent space that serves as an intermediate representation. To ground the framework theoretically, we derive a distillation loss using Young's Inequality, which decomposes the direct teacher-student distillation path into teacher-TA and TA-student dual paths, stabilizing optimization and strengthening knowledge transfer. Extensive experiments on the challenging nuScenes dataset demonstrate the effectiveness of our method, achieving an improvement of 4.2% mIoU over the Camera-only baseline, up to 45% higher than the improvement of other state-of-the-art KD methods.

Beomjun Kim, Kangyeon Kim, Sunwoo Kim et al.

AI safety has emerged as a critical priority as these systems are increasingly deployed in real-world applications. We propose the first domain-agnostic AI safety ensuring framework that achieves strong safety guarantees while preserving high performance, grounded in rigorous theoretical foundations. Our framework includes: (1) an optimization component with chance constraints, (2) a safety classification model, (3) internal test data, (4) conservative testing procedures, (5) informative dataset quality measures, and (6) continuous approximate loss functions with gradient computation. Furthermore, to our knowledge, we mathematically establish the first scaling law in AI safety research, relating data quantity to safety-performance trade-offs. Experiments across reinforcement learning, natural language generation, and production planning validate our framework and demonstrate superior performance. Notably, in reinforcement learning, we achieve 3 collisions during 10M actions, compared with 1,000-3,000 for PPO-Lag baselines at equivalent performance levels -- a safety level unattainable by previous AI methods. We believe our framework opens a new foundation for safe AI deployment across safety-critical domains.

Youngmin Oh, Beomjun Kim, Bumsub Ham

We address the problem of weakly-supervised semantic segmentation (WSSS) using bounding box annotations. Although object bounding boxes are good indicators to segment corresponding objects, they do not specify object boundaries, making it hard to train convolutional neural networks (CNNs) for semantic segmentation. We find that background regions are perceptually consistent in part within an image, and this can be leveraged to discriminate foreground and background regions inside object bounding boxes. To implement this idea, we propose a novel pooling method, dubbed background-aware pooling (BAP), that focuses more on aggregating foreground features inside the bounding boxes using attention maps. This allows to extract high-quality pseudo segmentation labels to train CNNs for semantic segmentation, but the labels still contain noise especially at object boundaries. To address this problem, we also introduce a noise-aware loss (NAL) that makes the networks less susceptible to incorrect labels. Experimental results demonstrate that learning with our pseudo labels already outperforms state-of-the-art weakly- and semi-supervised methods on the PASCAL VOC 2012 dataset, and the NAL further boosts the performance.

Beomjun Kim, Jean Ponce, Bumsub Ham

Joint image filters are used to transfer structural details from a guidance picture used as a prior to a target image, in tasks such as enhancing spatial resolution and suppressing noise. Previous methods based on convolutional neural networks (CNNs) combine nonlinear activations of spatially-invariant kernels to estimate structural details and regress the filtering result. In this paper, we instead learn explicitly sparse and spatially-variant kernels. We propose a CNN architecture and its efficient implementation, called the deformable kernel network (DKN), that outputs sets of neighbors and the corresponding weights adaptively for each pixel. The filtering result is then computed as a weighted average. We also propose a fast version of DKN that runs about seventeen times faster for an image of size 640 x 480. We demonstrate the effectiveness and flexibility of our models on the tasks of depth map upsampling, saliency map upsampling, cross-modality image restoration, texture removal, and semantic segmentation. In particular, we show that the weighted averaging process with sparsely sampled 3 x 3 kernels outperforms the state of the art by a significant margin in all cases.

Beomjun Kim, Jean Ponce, Bumsub Ham

We address the problem of upsampling a low-resolution (LR) depth map using a registered high-resolution (HR) color image of the same scene. Previous methods based on convolutional neural networks (CNNs) combine nonlinear activations of spatially-invariant kernels to estimate structural details from LR depth and HR color images, and regress upsampling results directly from the networks. In this paper, we revisit the weighted averaging process that has been widely used to transfer structural details from hand-crafted visual features to LR depth maps. We instead learn explicitly sparse and spatially-variant kernels for this task. To this end, we propose a CNN architecture and its efficient implementation, called the deformable kernel network (DKN), that outputs sparse sets of neighbors and the corresponding weights adaptively for each pixel. We also propose a fast version of DKN (FDKN) that runs about 17 times faster (0.01 seconds for a HR image of size 640 x 480). Experimental results on standard benchmarks demonstrate the effectiveness of our approach. In particular, we show that the weighted averaging process with 3 x 3 kernels (i.e., aggregating 9 samples sparsely chosen) outperforms the state of the art by a significant margin.