arXiv subjects

Sangho Lee

Publications and source records attributed to Sangho Lee.

At least 19 recordsLinked to original sources

arXiv2026

GARAGE: Characterizing the Automation Boundary in LLM-based Attack Graph Generation

While modern vehicle security depends on effective Cyber Threat Intelligence (CTI) synthesis, current automated tools struggle with unstructured data and automotive-specific architectural nuances. To bridge this gap, we introduce GARAGE, a RAG-powered framework that converts fragmented CTI into an actionable, domain-specific knowledge base for automated attack graph generation. GARAGE synthesizes a dataset of 12,786 CVEs and 140 incident reports into a STIX 2.1 and Auto-ISAC ATM-compliant knowledge base. By formalizing tactical-pattern-level scenarios through granular kill chain analysis, GARAGE achieves threat generation capabilities. Our 320 Leave-One-Out experiments reveal that the framework can accurately transfer security knowledge to entirely unseen vehicle architectures. Furthermore, we position GARAGE as a scalable TARA support tool within human-in-the-loop workflows, offering a comprehensive cost-performance analysis to guide its deployment across various LLM tiers.

cs.CR↗

arXiv2026

Molmo2: Open Weights and Data for Vision-Language Models with Video Understanding and Grounding

Today's strongest video-language models (VLMs) remain proprietary. The strongest open-weight models either rely on synthetic data from proprietary VLMs, effectively distilling from them, or do not disclose their training data or recipe. As a result, the open-source community lacks the foundations needed to improve on the state-of-the-art video (and image) language models. Crucially, many downstream applications require more than just high-level video understanding; they require grounding -- either by pointing or by tracking in pixels. Even proprietary models lack this capability. We present Molmo2, a new family of VLMs that are state-of-the-art among open-source models and demonstrate exceptional new capabilities in point-driven grounding in single image, multi-image, and video tasks. Our key contribution is a collection of 7 new video datasets and 2 multi-image datasets, including a dataset of highly detailed video captions for pre-training, a free-form video Q&A dataset for fine-tuning, a new object tracking dataset with complex queries, and an innovative new video pointing dataset, all collected without the use of closed VLMs. We also present a training recipe for this data utilizing an efficient packing and message-tree encoding scheme, and show bi-directional attention on vision tokens and a novel token-weight strategy improves performance. Our best-in-class 8B model outperforms others in the class of open weight and data models on short videos, counting, and captioning, and is competitive on long-videos. On video-grounding Molmo2 significantly outperforms existing open-weight models like Qwen3-VL (35.5 vs 29.6 accuracy on video counting) and surpasses proprietary models like Gemini 3 Pro on some tasks (38.4 vs 20.0 F1 on video pointing and 56.2 vs 41.1 J&F on video tracking).

cs.CV↗

arXiv2026

MolmoPoint: Better Pointing for VLMs with Grounding Tokens

Grounding has become a fundamental capability of vision-language models (VLMs). Most existing VLMs point by generating coordinates as part of their text output, which requires learning a complicated coordinate system and results in a high token count. Instead, we propose a more intuitive pointing mechanism that directly selects the visual tokens that contain the target concept. Our model generates a special pointing token that cross-attends to the input image or video tokens and selects the appropriate one. To make this model more fine-grained, we follow these pointing tokens with an additional special token that selects a fine-grained subpatch within the initially selected region, and then a third token that specifies a location within that subpatch. We further show that performance improves by generating points sequentially in a consistent order, encoding the relative position of the previously selected point, and including a special no-more-points class when selecting visual tokens. Using this method, we set a new state-of-the-art on image pointing (70.7% on PointBench), set a new state-of-the-art among fully open models on GUI pointing (61.1% on ScreenSpotPro), and improve video pointing (59.1% human preference win rate vs. a text coordinate baseline) and tracking (+6.3% gain on Molmo2Track). We additionally show that our method achieves much higher sample efficiency and discuss the qualitative differences that emerge from this design change.

cs.CV↗

arXiv2026

SAGE: Training Smart Any-Horizon Agents for Long Video Reasoning with Reinforcement Learning

As humans, we are natural any-horizon reasoners, i.e., we can decide whether to iteratively skim long videos or watch short ones in full when necessary for a given task. With this in mind, one would expect video reasoning models to reason flexibly across different durations. However, SOTA models are still trained to predict answers in a single turn while processing a large number of frames, akin to watching an entire long video, requiring significant resources. This raises the question: Is it possible to develop performant any-horizon video reasoning systems? Inspired by human behavior, we first propose SAGE, an agent system that performs multi-turn reasoning on long videos while handling simpler problems in a single turn. Secondly, we introduce an easy synthetic data generation pipeline using Gemini-2.5-Flash to train the orchestrator, SAGE-MM, which lies at the core of SAGE. We further propose an effective RL post-training recipe essential for instilling any-horizon reasoning ability in SAGE-MM. Thirdly, we curate SAGE-Bench with an average duration of greater than 700 seconds for evaluating video reasoning ability in real-world entertainment use cases. Lastly, we empirically validate the effectiveness of our system, data, and RL recipe, observing notable improvements of up to 6.1% on open-ended video reasoning tasks, as well as an impressive 8.2% improvement on videos longer than 10 minutes.

cs.CV↗

arXiv2026

Unified Spatio-Temporal Token Scoring for Efficient Video VLMs

Token pruning is essential for enhancing the computational efficiency of vision-language models (VLMs), particularly for video-based tasks where temporal redundancy is prevalent. Prior approaches typically prune tokens either (1) within the vision transformer (ViT) exclusively for unimodal perception tasks such as action recognition and object segmentation, without adapting to downstream vision-language tasks; or (2) only within the LLM while leaving the ViT output intact, often requiring complex text-conditioned token selection mechanisms. In this paper, we introduce Spatio-Temporal Token Scoring (STTS), a simple and lightweight module that prunes vision tokens across both the ViT and the LLM without text conditioning or token merging, and is fully compatible with end-to-end training. By learning how to score temporally via an auxiliary loss and spatially via LLM downstream gradients, aided by our efficient packing algorithm, STTS prunes 50% of vision tokens throughout the entire architecture, resulting in a 62% improvement in efficiency during both training and inference with only a 0.7% drop in average performance across 13 short and long video QA tasks. Efficiency gains increase with more sampled frames per video. Applying test-time scaling for long-video QA further yields performance gains of 0.5-1% compared to the baseline. Overall, STTS represents a novel, simple yet effective technique for unified, architecture-wide vision token pruning.

cs.CV↗

arXiv2025

MolmoAct: Action Reasoning Models that can Reason in Space

Reasoning is central to purposeful action, yet most robotic foundation models map perception and instructions directly to control, which limits adaptability, generalization, and semantic grounding. We introduce Action Reasoning Models (ARMs), a class of robotic foundation models that integrate perception, planning, and control through a structured three-stage pipeline. Our model, MolmoAct, encodes observations and instructions into depth-aware perception tokens, generates mid-level spatial plans as editable trajectory traces, and predicts precise low-level actions, enabling explainable and steerable behavior. MolmoAct-7B-D achieves strong performance across simulation and real-world settings: 70.5% zero-shot accuracy on SimplerEnv Visual Matching tasks, surpassing closed-source Pi-0 and GR00T N1.5; 86.6% average success on LIBERO, including an additional 6.3% gain over ThinkAct on long-horizon tasks; and in real-world fine-tuning, an additional 10% (single-arm) and an additional 22.7% (bimanual) task progression over Pi-0-FAST. It also outperforms baselines by an additional 23.3% on out-of-distribution generalization and achieves top human-preference scores for open-ended instruction following and trajectory steering. Furthermore, we release, for the first time, the MolmoAct Dataset -- a mid-training robot dataset comprising over 10,000 high quality robot trajectories across diverse scenarios and tasks. Training with this dataset yields an average 5.5% improvement in general performance over the base model. We release all model weights, training code, our collected dataset, and our action reasoning dataset, establishing MolmoAct as both a state-of-the-art robotics foundation model and an open blueprint for building ARMs that transform perception into purposeful action through structured reasoning. Blogpost: https://allenai.org/blog/molmoact

cs.RO↗

arXiv2025

One Diffusion to Generate Them All

We introduce OneDiffusion, a versatile, large-scale diffusion model that seamlessly supports bidirectional image synthesis and understanding across diverse tasks. It enables conditional generation from inputs such as text, depth, pose, layout, and semantic maps, while also handling tasks like image deblurring, upscaling, and reverse processes such as depth estimation and segmentation. Additionally, OneDiffusion allows for multi-view generation, camera pose estimation, and instant personalization using sequential image inputs. Our model takes a straightforward yet effective approach by treating all tasks as frame sequences with varying noise scales during training, allowing any frame to act as a conditioning image at inference time. Our unified training framework removes the need for specialized architectures, supports scalable multi-task training, and adapts smoothly to any resolution, enhancing both generalization and scalability. Experimental results demonstrate competitive performance across tasks in both generation and prediction such as text-to-image, multiview generation, ID preservation, depth estimation and camera pose estimation despite relatively small training dataset. Our code and checkpoint are freely available at https://github.com/lehduong/OneDiffusion

cs.CV↗

arXiv2025

ReSpec: Relevance and Specificity Grounded Online Filtering for Learning on Video-Text Data Streams

The rapid growth of video-text data presents challenges in storage and computation during training. Online learning, which processes streaming data in real-time, offers a promising solution to these issues while also allowing swift adaptations in scenarios demanding real-time responsiveness. One strategy to enhance the efficiency and effectiveness of learning involves identifying and prioritizing data that enhances performance on target downstream tasks. We propose Relevance and Specificity-based online filtering framework (ReSpec) that selects data based on four criteria: (i) modality alignment for clean data, (ii) task relevance for target focused data, (iii) specificity for informative and detailed data, and (iv) efficiency for low-latency processing. Relevance is determined by the probabilistic alignment of incoming data with downstream tasks, while specificity employs the distance to a root embedding representing the least specific data as an efficient proxy for informativeness. By establishing reference points from target task data, ReSpec filters incoming data in real-time, eliminating the need for extensive storage and compute. Evaluating on large-scale datasets WebVid2M and VideoCC3M, ReSpec attains state-of-the-art performance on five zeroshot video retrieval tasks, using as little as 5% of the data while incurring minimal compute. The source code is available at https://github.com/cdjkim/ReSpec.

cs.CV↗

arXiv2025

MAMS: Model-Agnostic Module Selection Framework for Video Captioning

Multi-modal transformers are rapidly gaining attention in video captioning tasks. Existing multi-modal video captioning methods typically extract a fixed number of frames, which raises critical challenges. When a limited number of frames are extracted, important frames with essential information for caption generation may be missed. Conversely, extracting an excessive number of frames includes consecutive frames, potentially causing redundancy in visual tokens extracted from consecutive video frames. To extract an appropriate number of frames for each video, this paper proposes the first model-agnostic module selection framework in video captioning that has two main functions: (1) selecting a caption generation module with an appropriate size based on visual tokens extracted from video frames, and (2) constructing subsets of visual tokens for the selected caption generation module. Furthermore, we propose a new adaptive attention masking scheme that enhances attention on important visual tokens. Our experiments on three different benchmark datasets demonstrate that the proposed framework significantly improves the performance of three recent video captioning models.

cs.CV↗

arXiv2024

Molmo and PixMo: Open Weights and Open Data for State-of-the-Art Vision-Language Models

Today's most advanced vision-language models (VLMs) remain proprietary. The strongest open-weight models rely heavily on synthetic data from proprietary VLMs to achieve good performance, effectively distilling these closed VLMs into open ones. As a result, the community has been missing foundational knowledge about how to build performant VLMs from scratch. We present Molmo, a new family of VLMs that are state-of-the-art in their class of openness. Our key contribution is a collection of new datasets called PixMo, including a dataset of highly detailed image captions for pre-training, a free-form image Q&A dataset for fine-tuning, and an innovative 2D pointing dataset, all collected without the use of external VLMs. The success of our approach relies on careful modeling choices, a well-tuned training pipeline, and, most critically, the quality of our newly collected datasets. Our best-in-class 72B model not only outperforms others in the class of open weight and data models, but also outperforms larger proprietary models including Claude 3.5 Sonnet, and Gemini 1.5 Pro and Flash, second only to GPT-4o based on both academic benchmarks and on a large human evaluation. Our model weights, new datasets, and source code are available at https://molmo.allenai.org/blog.

cs.CV↗

arXiv2024

TEEMATE: Fast and Efficient Confidential Container using Shared Enclave

Confidential container is becoming increasingly popular as it meets both needs for efficient resource management by cloud providers, and data protection by cloud users. Specifically, confidential containers integrate the container and the enclave, aiming to inherit the design-wise advantages of both (i.e., resource management and data protection). However, current confidential containers suffer from large performance overheads caused by i) a larger startup latency due to the enclave creation, and ii) a larger memory footprint due to the non-shareable characteristics of enclave memory. This paper explores a design conundrum of confidential container, examining why the confidential containers impose such large performance overheads. Surprisingly, we found there is a universal misconception that an enclave can only be used by a single (containerized) process that created it. However, an enclave can be shared across multiple processes, because an enclave is merely a set of physical resources while the process is an abstraction constructed by the host kernel. To this end, we introduce TeeMate, a new approach to utilize the enclaves on the host system. Especially, TeeMate designs the primitives to i) share the enclave memory between processes, thus preserving memory abstraction, and ii) assign the threads in enclave between processes, thus preserving thread abstraction. We concretized TeeMate on Intel SGX, and implemented confidential serverless computing and confidential database on top of TeeMate based confidential containers. The evaluation clearly demonstrated the strong practical impact of TeeMate by achieving at least 4.5 times lower latency and 2.8 times lower memory usage compared to the applications built on the conventional confidential containers.

cs.CR↗

arXiv2024

Finding NeMo: Negative-mined Mosaic Augmentation for Referring Image Segmentation

Referring Image Segmentation is a comprehensive task to segment an object referred by a textual query from an image. In nature, the level of difficulty in this task is affected by the existence of similar objects and the complexity of the referring expression. Recent RIS models still show a significant performance gap between easy and hard scenarios. We pose that the bottleneck exists in the data, and propose a simple but powerful data augmentation method, Negative-mined Mosaic Augmentation (NeMo). This method augments a training image into a mosaic with three other negative images carefully curated by a pretrained multimodal alignment model, e.g., CLIP, to make the sample more challenging. We discover that it is critical to properly adjust the difficulty level, neither too ambiguous nor too trivial. The augmented training data encourages the RIS model to recognize subtle differences and relationships between similar visual entities and to concretely understand the whole expression to locate the right target better. Our approach shows consistent improvements on various datasets and models, verified by extensive experiments.

cs.CV↗

arXiv2024

Can Language Models Laugh at YouTube Short-form Videos?

As short-form funny videos on social networks are gaining popularity, it becomes demanding for AI models to understand them for better communication with humans. Unfortunately, previous video humor datasets target specific domains, such as speeches or sitcoms, and mostly focus on verbal cues. We curate a user-generated dataset of 10K multimodal funny videos from YouTube, called ExFunTube. Using a video filtering pipeline with GPT-3.5, we verify both verbal and visual elements contributing to humor. After filtering, we annotate each video with timestamps and text explanations for funny moments. Our ExFunTube is unique over existing datasets in that our videos cover a wide range of domains with various types of humor that necessitate a multimodal understanding of the content. Also, we develop a zero-shot video-to-text prompting to maximize video humor understanding of large language models (LLMs). With three different evaluation methods using automatic scores, rationale quality experiments, and human evaluations, we show that our prompting significantly improves LLMs' ability for humor explanation.

cs.CL↗

arXiv2024

X-CANIDS: Signal-Aware Explainable Intrusion Detection System for Controller Area Network-Based In-Vehicle Network

Controller Area Network (CAN) is an essential networking protocol that connects multiple electronic control units (ECUs) in a vehicle. However, CAN-based in-vehicle networks (IVNs) face security risks owing to the CAN mechanisms. An adversary can sabotage a vehicle by leveraging the security risks if they can access the CAN bus. Thus, recent actions and cybersecurity regulations (e.g., UNR 155) require carmakers to implement intrusion detection systems (IDSs) in their vehicles. The IDS should detect cyberattacks and provide additional information to analyze conducted attacks. Although many IDSs have been proposed, considerations regarding their feasibility and explainability remain lacking. This study proposes X-CANIDS, which is a novel IDS for CAN-based IVNs. X-CANIDS dissects the payloads in CAN messages into human-understandable signals using a CAN database. The signals improve the intrusion detection performance compared with the use of bit representations of raw payloads. These signals also enable an understanding of which signal or ECU is under attack. X-CANIDS can detect zero-day attacks because it does not require any labeled dataset in the training phase. We confirmed the feasibility of the proposed method through a benchmark test on an automotive-grade embedded device with a GPU. The results of this work will be valuable to carmakers and researchers considering the installation of in-vehicle IDSs for their vehicles.

cs.CR↗

arXiv2023

Unified-IO 2: Scaling Autoregressive Multimodal Models with Vision, Language, Audio, and Action

We present Unified-IO 2, the first autoregressive multimodal model that is capable of understanding and generating image, text, audio, and action. To unify different modalities, we tokenize inputs and outputs -- images, text, audio, action, bounding boxes, etc., into a shared semantic space and then process them with a single encoder-decoder transformer model. Since training with such diverse modalities is challenging, we propose various architectural improvements to stabilize model training. We train our model from scratch on a large multimodal pre-training corpus from diverse sources with a multimodal mixture of denoisers objective. To learn an expansive set of skills, such as following multimodal instructions, we construct and finetune on an ensemble of 120 datasets with prompts and augmentations. With a single unified model, Unified-IO 2 achieves state-of-the-art performance on the GRIT benchmark and strong results in more than 35 benchmarks, including image generation and understanding, natural language understanding, video and audio understanding, and robotic manipulation. We release all our models to the research community.

cs.CV↗

arXiv2023

Integrated Path Tracking with DYC and MPC using LSTM Based Tire Force Estimator for Four-wheel Independent Steering and Driving Vehicle

Active collision avoidance system plays a crucial role in ensuring the lateral safety of autonomous vehicles, and it is primarily related to path planning and tracking control algorithms. In particular, the direct yaw-moment control (DYC) system can significantly improve the lateral stability of a vehicle in environments with sudden changes in road conditions. In order to apply the DYC algorithm, it is very important to accurately consider the properties of tire forces with complex nonlinearity for control to ensure the lateral stability of the vehicle. In this study, longitudinal and lateral tire forces for safety path tracking were simultaneously estimated using a long short-term memory (LSTM) neural network based estimator. Furthermore, to improve path tracking performance in case of sudden changes in road conditions, a system has been developed by combining 4-wheel independent steering (4WIS) model predictive control (MPC) and 4-wheel independent drive (4WID) direct yaw-moment control (DYC). The estimation performance of the extended Kalman filter (EKF), which are commonly used for tire force estimation, was compared. In addition, the estimated longitudinal and lateral tire forces of each wheel were applied to the proposed system, and system verification was performed through simulation using a vehicle dynamics simulator. Consequently, the proposed method, the integrated path tracking algorithm with DYC and MPC using the LSTM based estimator, was validated to significantly improve the vehicle stability in suddenly changing road conditions.

cs.RO↗

arXiv2023

Seamless monolithic three-dimensional integration of single-crystalline films by growth

The demand for the three-dimensional (3D) integration of electronic components is on a steady rise. The through-silicon-via (TSV) technique emerges as the only viable method for integrating single-crystalline device components in a 3D format, despite encountering significant processing challenges. While monolithic 3D (M3D) integration schemes show promise, the seamless connection of single-crystalline semiconductors without intervening wafers has yet to be demonstrated. This challenge arises from the inherent difficulty of growing single crystals on amorphous or polycrystalline surfaces post the back-end-of-the-line process at low temperatures to preserve the underlying circuitry. Consequently, a practical growth-based solution for M3D of single crystals remains elusive. Here, we present a method for growing single-crystalline channel materials, specifically composed of transition metal dichalcogenides, on amorphous and polycrystalline surfaces at temperatures lower than 400 °C. Building on this developed technique, we demonstrate the seamless monolithic integration of vertical single-crystalline logic transistor arrays. This accomplishment leads to the development of unprecedented vertical CMOS arrays, thereby constructing vertical inverters. Ultimately, this achievement sets the stage to pave the way for M3D integration of various electronic and optoelectronic hardware in the form of single crystals.

physics.app-ph↗

arXiv2022

Mitigating Low-volume DoS Attacks with Data-driven Resource Accounting

Low-volume Denial-of-Service (μDoS) attacks have been demonstrated to fundamentally bypass traditional DoS mitigation schemes based on the flow and volume of network packets. In this paper, we propose a data-driven approach, called ROKI, that accurately tracks internal resource utilization and allocation associated with each packet (or session), making it possible to tame resource exhaustion caused by μDoS attacks. Since ROKI focuses on capturing the symptom of DoS, it can effectively mitigate previously unknown μDoS attacks. To enable a finer-grain resource tracking, ROKI provided in concept the accounting capabilities to each packet itself, so we called data-driven: it monitors resource utilization at the link, network, transport layers in the kernel, as well as application layers, and attributes back to the associated packet. Given the resource usages of each packet, ROKI can reclaim (or prevent) the system resources from malicious packets (or attackers) whenever it encounters system-wide resource exhaustion. To provide lightweight resource tracking, ROKI carefully multiplexes hardware performance counters whenever necessary. Our evaluation shows that ROKI's approach is indeed effective in mitigating real-world μDoS attacks with negligible performance overheads - incurring 3%-4% throughput and latency overheads on average when the system is throttled.

cs.CR↗