arXiv subjects

Seung-Hoon Na

Publications and source records attributed to Seung-Hoon Na.

6 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

ROSAQ: Rotation-based Saliency-Aware Weight Quantization for Efficiently Compressing Large Language Models

Quantization has been widely studied as an effective technique for reducing the memory requirement of large language models (LLMs), potentially improving the latency time as well. Utilizing the characteristic of rotational invariance of transformer, we propose the rotation-based saliency-aware weight quantization (ROSAQ), which identifies salient channels in the projection feature space, not in the original feature space, where the projected "principal" dimensions are naturally considered as "salient" features. The proposed ROSAQ consists of 1) PCA-based projection, which first performs principal component analysis (PCA) on a calibration set and transforms via the PCA projection, 2) Salient channel dentification, which selects dimensions corresponding to the K-largest eigenvalues as salient channels, and 3) Saliency-aware quantization with mixed-precision, which uses FP16 for salient dimensions and INT3/4 for other dimensions. Experiment results show that ROSAQ shows improvements over the baseline saliency-aware quantization on the original feature space and other existing quantization methods. With kernel fusion, ROSAQ presents about 2.3x speed up over FP16 implementation in generating 256 tokens with a batch size of 64.

cs.CL↗

arXiv2025

CacheFocus: Dynamic Cache Re-Positioning for Efficient Retrieval-Augmented Generation

Large Language Models (LLMs) excel across a variety of language tasks yet are constrained by limited input lengths and high computational costs. Existing approaches\textemdash such as relative positional encodings (e.g., RoPE, ALiBi) and sliding window mechanisms\textemdash partially alleviate these issues but often require additional training or suffer from performance degradation with longer inputs. In this paper, we introduce \textbf{\textit{CacheFocus}}, a method that enhances length normalization and reduces inference latency without any further training. Our approach leverages query-independent, offline caching to efficiently reuse a Context KV Cache Store. We address the amplification of abnormal token distributions problem by re-positioning cached keys and introducing Layer-Adaptive Cache Pruning to discard low-relevance caches during pre-filling. Additionally, our Adaptive Positional Allocation Strategy dynamically reassigns cache positions to maximize the use of the available positional encoding range. Experiments on the Natural Questions and TriviaQA datasets demonstrate that CacheFocus outperforms alternative methods even when inputs exceed the $4$K limit of the \texttt{LLaMA-2} model, emphasizing its practical effectiveness for long-context LLMs. Moreover, even with large maximum input length of \texttt{Qwen2}, the performance of CacheFocus shows that it maintains consistent performance even as the number of documents increases, effectively managing long-text generation without degradation.

cs.CL↗

arXiv2023

Feature Structure Distillation with Centered Kernel Alignment in BERT Transferring

Knowledge distillation is an approach to transfer information on representations from a teacher to a student by reducing their difference. A challenge of this approach is to reduce the flexibility of the student's representations inducing inaccurate learning of the teacher's knowledge. To resolve it in transferring, we investigate distillation of structures of representations specified to three types: intra-feature, local inter-feature, global inter-feature structures. To transfer them, we introduce feature structure distillation methods based on the Centered Kernel Alignment, which assigns a consistent value to similar features structures and reveals more informative relations. In particular, a memory-augmented transfer method with clustering is implemented for the global structures. The methods are empirically analyzed on the nine tasks for language understanding of the GLUE dataset with Bidirectional Encoder Representations from Transformers (BERT), which is a representative neural language model. In the results, the proposed methods effectively transfer the three types of structures and improve performance compared to state-of-the-art distillation methods. Indeed, the code for the methods is available in https://github.com/maroo-sky/FSD.

cs.CL↗

arXiv2015

Two-Stage Document Length Normalization for Information Retrieval

The standard approach for term frequency normalization is based only on the document length. However, it does not distinguish the verbosity from the scope, these being the two main factors determining the document length. Because the verbosity and scope have largely different effects on the increase in term frequency, the standard approach can easily suffer from insufficient or excessive penalization depending on the specific type of long document. To overcome these problems, this paper proposes two-stage normalization by performing verbosity and scope normalization separately, and by employing different penalization functions. In verbosity normalization, each document is pre-normalized by dividing the term frequency by the verbosity of the document. In scope normalization, an existing retrieval model is applied in a straightforward manner to the pre-normalized document, finally leading us to formulate our proposed verbosity normalized (VN) retrieval model. Experimental results carried out on standard TREC collections demonstrate that the VN model leads to marginal but statistically significant improvements over standard retrieval models.

cs.IR↗

arXiv2015

Improving Term Frequency Normalization for Multi-topical Documents, and Application to Language Modeling Approaches

Term frequency normalization is a serious issue since lengths of documents are various. Generally, documents become long due to two different reasons - verbosity and multi-topicality. First, verbosity means that the same topic is repeatedly mentioned by terms related to the topic, so that term frequency is more increased than the well-summarized one. Second, multi-topicality indicates that a document has a broad discussion of multi-topics, rather than single topic. Although these document characteristics should be differently handled, all previous methods of term frequency normalization have ignored these differences and have used a simplified length-driven approach which decreases the term frequency by only the length of a document, causing an unreasonable penalization. To attack this problem, we propose a novel TF normalization method which is a type of partially-axiomatic approach. We first formulate two formal constraints that the retrieval model should satisfy for documents having verbose and multi-topicality characteristic, respectively. Then, we modify language modeling approaches to better satisfy these two constraints, and derive novel smoothing methods. Experimental results show that the proposed method increases significantly the precision for keyword queries, and substantially improves MAP (Mean Average Precision) for verbose queries.

cs.IR↗