arXiv subjects

Sungjae Lee

Publications and source records attributed to Sungjae Lee.

At least 19 recordsLinked to original sources

arXiv2026

PaT: Planning-after-Trial for Efficient Test-Time Code Generation

Beyond training-time optimization, scaling test-time computation has emerged as a key paradigm to extend the reasoning capabilities of Large Language Models (LLMs). However, most existing methods adopt a rigid Planning-before-Trial (PbT) policy, which inefficiently allocates test-time compute by incurring planning overhead even on directly solvable problems. We propose Planning-after-Trial (PaT), an adaptive policy for code generation that invokes a planner only upon verification failure. This adaptive policy naturally enables a heterogeneous model configuration: a cost-efficient model handles generation attempts, while a powerful model is reserved for targeted planning interventions. Empirically, across multiple benchmarks and model families, our approach significantly advances the cost-performance Pareto frontier. Notably, our heterogeneous configuration achieves performance comparable to a large homogeneous model while reducing inference cost by approximately 69\%.

cs.CL↗

arXiv2026

Physically Guided Visual Mass Estimation from a Single RGB Image

Estimating object mass from visual input is challenging because mass depends jointly on geometric volume and material-dependent density, neither of which is directly observable from RGB appearance. Consequently, mass prediction from pixels is ill-posed and therefore benefits from physically meaningful representations to constrain the space of plausible solutions. We propose a physically structured framework for single-image mass estimation that addresses this ambiguity by aligning visual cues with the physical factors governing mass. From a single RGB image, we recover object-centric three-dimensional geometry via monocular depth estimation to inform volume and extract coarse material semantics using a vision-language model to guide density-related reasoning. These geometry, semantic, and appearance representations are fused through an instance-adaptive gating mechanism, and two physically guided latent factors (volume- and density-related) are predicted through separate regression heads under mass-only supervision. Experiments on image2mass and ABO-500 show that the proposed method consistently outperforms state-of-the-art methods.

cs.CV↗

arXiv2026

ICaRus: Identical Cache Reuse for Efficient Multi Model Inference

Multi model inference has recently emerged as a prominent paradigm, particularly in the development of agentic AI systems. However, in such scenarios, each model must maintain its own Key-Value (KV) cache for the identical prompt, leading to substantial memory consumption. This explosive growth of KV caches forces LLM serving systems to evict previously stored caches, which in turn introduces significant recomputation overhead whenever the evicted caches are required again. Moreover, prefix caching is inherently infeasible across different models, forcing each model to recompute KV cache for the identical prompt, which leads to significant overhead. To alleviate these issues, we propose Identical Cache Reuse (ICaRus), a novel architecture that allows multiple models to share identical KV caches across all layers. ICaRus is based on the key observation that a decoder-only Transformer can be conceptually decomposed into a logical encoder, which generates KV caches, and a logical decoder, which predicts output tokens from the KV caches. ICaRus fine-tunes only the logical decoder while freezing the logical encoder, enabling multiple models to share an identical KV cache. This eliminates cache memory explosion and unexpected evictions while also allowing cross-model reuse of KV caches for new input tokens, thereby removing redundant recomputation in multi model inference achieving both efficiency and scalability. Moreover, by incorporating lightweight adapters such as LoRA, ICaRus parallelizes KV cache generation and next-token prediction during decoding. ICaRus achieves comparable accuracy to task-specific fine-tuned model across a diverse set of tasks, while allowing multiple specialized models to fully share KV caches. ICaRus achieves up to 11.1x lower P95 latency and 3.8x higher throughput in multi agent workflow with 8 different models, compared to conventional multi model system.

cs.LG↗

arXiv2026

PrefillShare: A Shared Prefill Module for KV Reuse in Multi-LLM Disaggregated Serving

Multi-agent systems increasingly orchestrate multiple specialized language models to solve complex real-world problems, often invoking them over a shared context. This execution pattern repeatedly processes the same prompt prefix across models. Consequently, each model redundantly executes the prefill stage and maintains its own key-value (KV) cache, increasing aggregate prefill load and worsening tail latency by intensifying prefill-decode interference in existing LLM serving stacks. Disaggregated serving reduces such interference by placing prefill and decode on separate GPUs, but disaggregation does not fundamentally eliminate inter-model redundancy in computation and KV storage for the same prompt. To address this issue, we propose PrefillShare, a novel algorithm that enables sharing the prefill stage across multiple models in a disaggregated setting. PrefillShare factorizes the model into prefill and decode modules, freezes the prefill module, and fine-tunes only the decode module. This design allows multiple task-specific models to share a prefill module and the KV cache generated for the same prompt. We further introduce a routing mechanism that enables effective prefill sharing across heterogeneous models in a vLLM-based disaggregated system. PrefillShare not only matches full fine-tuning accuracy on a broad range of tasks and models, but also delivers 4.5x lower p95 latency and 3.9x higher throughput in multi-model agent workloads.

cs.LG↗

arXiv2025

SelfJudge: Faster Speculative Decoding via Self-Supervised Judge Verification

Speculative decoding accelerates LLM inference by verifying candidate tokens from a draft model against a larger target model. Recent judge decoding boosts this process by relaxing verification criteria by accepting draft tokens that may exhibit minor discrepancies from target model output, but existing methods are restricted by their reliance on human annotations or tasks with verifiable ground truths, limiting generalizability across diverse NLP tasks. We propose SelfJudge, which trains judge verifiers via self-supervision of the target model. Our method measures semantic preservation by assessing whether token-substituted responses preserve the meaning of original responses, enabling automatic verifier training across diverse NLP tasks. Our experiments show SelfJudge achieves superior inference-accuracy trade-offs than judge decoding baselines, offering a broadly applicable solution for faster LLM inference.

cs.CL↗

arXiv2025

CANVAS: Commonsense-Aware Navigation System for Intuitive Human-Robot Interaction

Real-life robot navigation involves more than just reaching a destination; it requires optimizing movements while addressing scenario-specific goals. An intuitive way for humans to express these goals is through abstract cues like verbal commands or rough sketches. Such human guidance may lack details or be noisy. Nonetheless, we expect robots to navigate as intended. For robots to interpret and execute these abstract instructions in line with human expectations, they must share a common understanding of basic navigation concepts with humans. To this end, we introduce CANVAS, a novel framework that combines visual and linguistic instructions for commonsense-aware navigation. Its success is driven by imitation learning, enabling the robot to learn from human navigation behavior. We present COMMAND, a comprehensive dataset with human-annotated navigation results, spanning over 48 hours and 219 km, designed to train commonsense-aware navigation systems in simulated environments. Our experiments show that CANVAS outperforms the strong rule-based system ROS NavStack across all environments, demonstrating superior performance with noisy instructions. Notably, in the orchard environment, where ROS NavStack records a 0% total success rate, CANVAS achieves a total success rate of 67%. CANVAS also closely aligns with human demonstrations and commonsense constraints, even in unseen environments. Furthermore, real-world deployment of CANVAS showcases impressive Sim2Real transfer with a total success rate of 69%, highlighting the potential of learning from human demonstrations in simulated environments for real-world applications.

cs.RO↗

arXiv2025

Subspace-based Approximate Hessian Method for Zeroth-Order Optimization

Zeroth-order optimization addresses problems where gradient information is inaccessible or impractical to compute. While most existing methods rely on first-order approximations, incorporating second-order (curvature) information can, in principle, significantly accelerate convergence. However, the high cost of function evaluations required to estimate Hessian matrices often limits practical applicability. We present the subspace-based approximate Hessian (ZO-SAH) method, a zeroth-order optimization algorithm that mitigates these costs by focusing on randomly selected two-dimensional subspaces. Within each subspace, ZO-SAH estimates the Hessian by fitting a quadratic polynomial to the objective function and extracting its second-order coefficients. To further reduce function-query costs, ZO-SAH employs a periodic subspace-switching strategy that reuses function evaluations across optimization steps. Experiments on eight benchmark datasets, including logistic regression and deep neural network training tasks, demonstrate that ZO-SAH achieves significantly faster convergence than existing zeroth-order methods.

cs.LG↗

arXiv2025

Semantic Exploration with Adaptive Gating for Efficient Problem Solving with Language Models

Recent advancements in large language models (LLMs) have shown remarkable potential in various complex tasks requiring multi-step reasoning methods like tree search to explore diverse reasoning paths. However, existing methods often suffer from computational inefficiency and redundancy. First, they overlook the diversity of task difficulties, leading to unnecessarily extensive searches even for easy tasks. Second, they neglect the semantics of reasoning paths, resulting in redundant exploration of semantically identical paths. To address these limitations, we propose Semantic Exploration with Adaptive Gating (SEAG), a computationally efficient method. SEAG employs an adaptive gating mechanism that dynamically decides whether to conduct a tree search, based on the confidence level of answers from a preceding simple reasoning method. Furthermore, its tree-based exploration consolidates semantically identical reasoning steps, reducing redundant explorations while maintaining or even improving accuracy. Our extensive experiments demonstrate that SEAG significantly improves accuracy by 4.3% on average while requiring only 31% of computational costs compared to existing tree search-based methods on complex reasoning benchmarks including GSM8K and ARC with diverse language models such as Llama2, Llama3, and Mistral. Our code is available at https://github.com/ml-postech/SEAG-semantic-exploration-with-adaptive-gating .

cs.AI↗

arXiv2025

Efficient Latent Semantic Clustering for Scaling Test-Time Computation of LLMs

Scaling test-time computation--generating and analyzing multiple or sequential outputs for a single input--has become a promising strategy for improving the reliability and quality of large language models (LLMs), as evidenced by advances in uncertainty quantification and multi-step reasoning. A key shared component is semantic clustering, which groups outputs that differ in form but convey the same meaning. Semantic clustering enables estimation of the distribution over the semantics of outputs and helps avoid redundant exploration of reasoning paths. However, existing approaches typically rely on external models, which introduce substantial computational overhead and often fail to capture context-aware semantics. We propose Latent Semantic Clustering (LSC), a lightweight and context-sensitive method that leverages the generator LLM's internal hidden states for clustering, eliminating the need for external models. Our extensive experiment across various LLMs and datasets shows that LSC significantly improves the computational efficiency of test-time scaling while maintaining or exceeding the performance of existing methods.

cs.CL↗

arXiv2025

VisEscape: A Benchmark for Evaluating Exploration-driven Decision-making in Virtual Escape Rooms

Escape rooms present a unique cognitive challenge that demands exploration-driven planning: with the sole instruction to 'escape the room', players must actively search their environment, collecting information, and finding solutions through repeated trial and error. Motivated by this, we introduce VisEscape, a benchmark of 20 virtual escape rooms specifically designed to evaluate AI models under these challenging conditions, where success depends not only on solving isolated puzzles but also on iteratively constructing and refining spatial-temporal knowledge of a dynamically changing environment. On VisEscape, we observe that even state-of-the-art multi-modal models generally fail to escape the rooms, showing considerable variation in their progress and problem-solving approaches. We find that integrating memory management and reasoning contributes to efficient exploration and enables successive hypothesis formulation and testing, thereby leading to significant improvements in dynamic and exploration-driven environments

cs.AI↗

arXiv2025

Self-Training Large Language Models with Confident Reasoning

Large language models (LLMs) have shown impressive performance by generating reasoning paths before final answers, but learning such a reasoning path requires costly human supervision. To address this issue, recent studies have explored self-training methods that improve reasoning capabilities using pseudo-labels generated by the LLMs themselves. Among these, confidence-based self-training fine-tunes LLMs to prefer reasoning paths with high-confidence answers, where confidence is estimated via majority voting. However, such methods exclusively focus on the quality of the final answer and may ignore the quality of the reasoning paths, as even an incorrect reasoning path leads to a correct answer by chance. Instead, we advocate the use of reasoning-level confidence to identify high-quality reasoning paths for self-training, supported by our empirical observations. We then propose a new self-training method, CORE-PO, that fine-tunes LLMs to prefer high-COnfidence REasoning paths through Policy Optimization. Our experiments show that CORE-PO improves the accuracy of outputs on four in-distribution and two out-of-distribution benchmarks, compared to existing self-training methods.

cs.LG↗

arXiv2025

GraspCorrect: Robotic Grasp Correction via Vision-Language Model-Guided Feedback

Despite significant advancements in robotic manipulation, achieving consistent and stable grasping remains a fundamental challenge, often limiting the successful execution of complex tasks. Our analysis reveals that even state-of-the-art policy models frequently exhibit unstable grasping behaviors, leading to failure cases that create bottlenecks in real-world robotic applications. To address these challenges, we introduce GraspCorrect, a plug-and-play module designed to enhance grasp performance through vision-language model-guided feedback. GraspCorrect employs an iterative visual question-answering framework with two key components: grasp-guided prompting, which incorporates task-specific constraints, and object-aware sampling, which ensures the selection of physically feasible grasp candidates. By iteratively generating intermediate visual goals and translating them into joint-level actions, GraspCorrect significantly improves grasp stability and consistently enhances task success rates across existing policy models in the RLBench and CALVIN datasets.

cs.AI↗

arXiv2025

Piece of Table: A Divide-and-Conquer Approach for Selecting Subtables in Table Question Answering

Applying language models (LMs) to tables is challenging due to the inherent structural differences between two-dimensional tables and one-dimensional text for which the LMs were originally designed. Furthermore, when applying linearized tables to LMs, the maximum token lengths often imposed in self-attention calculations make it difficult to comprehensively understand the context spread across large tables. To address these challenges, we present PieTa (Piece of Table), a new framework for subtable-based question answering (QA). PieTa operates through an iterative process of dividing tables into smaller windows, using LMs to select relevant cells within each window, and merging these cells into a subtable. This multi-resolution approach captures dependencies across multiple rows and columns while avoiding the limitations caused by long context inputs. Instantiated as a simple iterative subtable union algorithm, PieTa demonstrates improved performance over previous subtable-based QA approaches.

cs.CL↗

arXiv2024

Selective Vision is the Challenge for Visual Reasoning: A Benchmark for Visual Argument Understanding

Visual arguments, often used in advertising or social causes, rely on images to persuade viewers to do or believe something. Understanding these arguments requires selective vision: only specific visual stimuli within an image are relevant to the argument, and relevance can only be understood within the context of a broader argumentative structure. While visual arguments are readily appreciated by human audiences, we ask: are today's AI capable of similar understanding? We present VisArgs, a dataset of 1,611 images annotated with 5,112 visual premises (with regions), 5,574 commonsense premises, and reasoning trees connecting them into structured arguments. We propose three tasks for evaluating visual argument understanding: premise localization, premise identification, and conclusion deduction. Experiments show that 1) machines struggle to capture visual cues: GPT-4-O achieved 78.5% accuracy, while humans reached 98.0%. Models also performed 19.5% worse when distinguishing between irrelevant objects within the image compared to external objects. 2) Providing relevant visual premises improved model performance significantly.

cs.CL↗

arXiv2024

LUT-GEMM: Quantized Matrix Multiplication based on LUTs for Efficient Inference in Large-Scale Generative Language Models

Recent advances in self-supervised learning and the Transformer architecture have significantly improved natural language processing (NLP), achieving remarkably low perplexity. However, the growing size of NLP models introduces a memory wall problem during the generation phase. To mitigate this issue, recent efforts have focused on quantizing model weights to sub-4-bit precision while preserving full precision for activations, resulting in practical speed-ups during inference on a single GPU. However, these improvements primarily stem from reduced memory movement, which necessitates a resource-intensive dequantization process rather than actual computational reduction. In this paper, we introduce LUT-GEMM, an efficient kernel for quantized matrix multiplication, which not only eliminates the resource-intensive dequantization process but also reduces computational costs compared to previous kernels for weight-only quantization. Furthermore, we proposed group-wise quantization to offer a flexible trade-off between compression ratio and accuracy. The impact of LUT-GEMM is facilitated by implementing high compression ratios through low-bit quantization and efficient LUT-based operations. We show experimentally that when applied to the OPT-175B model with 3-bit quantization, LUT-GEMM substantially accelerates token generation latency, achieving a remarkable 2.1$\times$ improvement on a single GPU when compared to OPTQ, which relies on the costly dequantization process.

cs.DC↗

arXiv2023

HPCClusterScape: Increasing Transparency and Efficiency of Shared High-Performance Computing Clusters for Large-scale AI Models

The emergence of large-scale AI models, like GPT-4, has significantly impacted academia and industry, driving the demand for high-performance computing (HPC) to accelerate workloads. To address this, we present HPCClusterScape, a visualization system that enhances the efficiency and transparency of shared HPC clusters for large-scale AI models. HPCClusterScape provides a comprehensive overview of system-level (e.g., partitions, hosts, and workload status) and application-level (e.g., identification of experiments and researchers) information, allowing HPC operators and machine learning researchers to monitor resource utilization and identify issues through customizable violation rules. The system includes diagnostic tools to investigate workload imbalances and synchronization bottlenecks in large-scale distributed deep learning experiments. Deployed in industrial-scale HPC clusters, HPCClusterScape incorporates user feedback and meets specific requirements. This paper outlines the challenges and prerequisites for efficient HPC operation, introduces the interactive visualization system, and highlights its contributions in addressing pain points and optimizing resource utilization in shared HPC clusters.

cs.HC↗

arXiv2022

PROFET: Profiling-based CNN Training Latency Prophet for GPU Cloud Instances

Training a Convolutional Neural Network (CNN) model typically requires significant computing power, and cloud computing resources are widely used as a training environment. However, it is difficult for CNN algorithm developers to keep up with system updates and apply them to their training environment due to quickly evolving cloud services. Thus, it is important for cloud computing service vendors to design and deliver an optimal training environment for various training tasks to lessen system operation management overhead of algorithm developers. To achieve the goal, we propose PROFET, which can predict the training latency of arbitrary CNN implementation on various Graphical Processing Unit (GPU) devices to develop a cost-effective and time-efficient training cloud environment. Different from the previous training latency prediction work, PROFET does not rely on the implementation details of the CNN architecture, and it is suitable for use in a public cloud environment. Thorough evaluations reveal the superior prediction accuracy of PROFET compared to the state-of-the-art related work, and the demonstration service presents the practicality of the proposed system.

cs.DC↗

arXiv2022

SpotLake: Diverse Spot Instance Dataset Archive Service

Public cloud service vendors provide a surplus of computing resources at a cheaper price as a spot instance. Despite the cheaper price, the spot instance can be forced to be shutdown at any moment whenever the surplus resources are in shortage. To enhance spot instance usage, vendors provide diverse spot instance datasets. Amon them, the spot price information has been most widely used so far. However, the tendency toward barely changing spot price weakens the applicability of the spot price dataset. Besides the price dataset, the recently introduced spot instance availability and interruption ratio datasets can help users better utilize spot instances, but they are rarely used in reality. With a thorough analysis, we could uncover major hurdles when using the new datasets concerning the lack of historical information, query constraints, and limited query interfaces. To overcome them, we develop SpotLake, a spot instance data archive web service that provides historical information of various spot instance datasets. Novel heuristics to collect various datasets and a data serving architecture are presented. Through real-world spot instance availability experiments, we present the applicability of the proposed system. SpotLake is publicly available as a web service to speed up cloud system research to improve spot instance usage and availability while reducing cost.

cs.DC↗