arXiv subjects

Donghyeok Koh

Publications and source records attributed to Donghyeok Koh.

3 recordsLinked to original sources

arXiv2026

Global Optimization and Inference-Time Region Grafting for Agentic Workflows

Recent advances in agentic workflow optimization automate workflow design through task-specific workflow search or input-conditioned architecture selection. However, they determine the workflow before execution and cannot adapt failed workflow regions using execution-time label-free quality signals. Naively enabling such inference-time adaptation through whole-workflow re-optimization would be computationally prohibitive. To tackle this challenge, we introduce GRAFT, which preserves a globally optimized workflow while locally replacing only selected regions for each input. Without parameter training, GRAFT evaluates region-level alternatives using label-free execution-quality signals and accepts only replacements that improve local quality while preserving workflow-level consistency, thereby enabling instance-wise adaptation without whole-workflow re-optimization. GRAFT applies without modification across a range of tasks spanning mathematical reasoning, code generation, and multi-hop and knowledge-intensive question answering. Under matched optimizer and executor settings, it improves over the strongest prior workflow-optimization method, MaAS, by 3.85 points on average. Replacing only the executor with a stronger model yields further gains without re-optimizing the global workflow. This suggests that an optimized workflow is not merely a static optimization artifact, but an adaptable execution policy that can evolve with inference-time feedback and stronger executors.

cs.CL↗

arXiv2025

TRAO Survey of the Nearby Filamentary Molecular Clouds, the Universal Nursery of Stars (TRAO-FUNS). IV. Filaments and Dense Cores in the W40 and Serpens South Regions of Aquila

We present the results of molecular line observations toward the W40 and Serpens South regions of the Aquila molecular cloud complex, conducted as part of the TRAO-FUNS project to investigate the role of filamentary structures in the formation of dense cores and stars in molecular clouds. We performed a Gaussian decomposition of the C$^{18}$O spectra to disentangle multiple velocity components along the line-of-sight and a `Friends-of-Friends' algorithm on these decomposed components to identify 24 velocity-coherent filaments in the observed region. The `FellWalker' algorithm is applied on the N$_{2}$H$^{+}$ integrated intensity map to identify the dense cores embedded within the filaments. Many of the filaments previously identified from the Herschel survey are found to contain multiple velocity-coherent filaments. Virial analysis indicated that all of our identified filaments are thermally supercritical and gravitationally bound. Velocity gradients are observed along the filaments in the vicinity of embedded dense cores, indicating the presence of longitudinal flows that contribute to core formation. The median mass flow rate across the observed region is estimated to be $\sim$35 M$_{\odot}$ Myr$^{-1}$, with Serpens South showing a rate $\sim$40\% higher than W40. The analysis of non-thermal motions revealed that the dense cores mainly show subsonic to transonic motions, while their host filaments are mostly supersonic, suggesting that the turbulent motions in filaments may dissipate on smaller scales, allowing core formation. These findings highlight the essential role of the filaments' criticality, mass flow, and turbulent dissipation in the formation of dense cores within the filaments.

astro-ph.GA↗

arXiv2025

ScholarBench: A Bilingual Benchmark for Abstraction, Comprehension, and Reasoning Evaluation in Academic Contexts

Prior benchmarks for evaluating the domain-specific knowledge of large language models (LLMs) lack the scalability to handle complex academic tasks. To address this, we introduce \texttt{ScholarBench}, a benchmark centered on deep expert knowledge and complex academic problem-solving, which evaluates the academic reasoning ability of LLMs and is constructed through a three-step process. \texttt{ScholarBench} targets more specialized and logically complex contexts derived from academic literature, encompassing five distinct problem types. Unlike prior benchmarks, \texttt{ScholarBench} evaluates the abstraction, comprehension, and reasoning capabilities of LLMs across eight distinct research domains. To ensure high-quality evaluation data, we define category-specific example attributes and design questions that are aligned with the characteristic research methodologies and discourse structures of each domain. Additionally, this benchmark operates as an English-Korean bilingual dataset, facilitating simultaneous evaluation for linguistic capabilities of LLMs in both languages. The benchmark comprises 5,031 examples in Korean and 5,309 in English, with even state-of-the-art models like o3-mini achieving an average evaluation score of only 0.543, demonstrating the challenging nature of this benchmark.

cs.CL↗