Searcharxiv⌕ Search

arXiv subjects

Gyuwan Kim

Publications and source records attributed to Gyuwan Kim.

At least 19 recordsLinked to original sources
arXiv2026

CoinRAG: Contextualized Information Nugget KV Cache Reuse for Long-Context RAG

Recent optimization studies on Retrieval-Augmented Generation (RAG) have exploited chunk-level KV cache reuse to avoid processing long retrieved contexts for higher efficiency, while significant information redundancy and noise still remain in the coarse-grained chunks. This paper optimizes the Pareto frontier under low prefill latency constraints while maximizing accuracy by proposing CoinRAG (Contextualized Information Nugget KV Cache Reuse for Long-Context RAG). The name metaphorically reflects our core mechanism: much like assembling small tokens (or "coins") to accumulate a larger value, CoinRAG compositionally reuses offline-computed, fine-grained nugget caches to form a learned contextual representation efficiently in a more semantically relevant but compact manner. Specifically, instead of full-chunk encoding, CoinRAG identifies query-relevant semantic units within retrieved chunks through two-stage retrieval and seamlessly assembles their sliced KV representations with a chunk-level context. Extensive evaluations on LongBench multi-hop question answering tasks demonstrate that CoinRAG significantly reduces operational costs and outperforms the other baselines with a new Pareto frontier and an average 5.3% relative improvement in answer quality (F1) under a standard fast prefill latency budget.

cs.CL↗
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↗
arXiv2026

KnapSpec: Self-Speculative Decoding via Adaptive Layer Selection as a Knapsack Problem

Self-speculative decoding (SSD) accelerates LLM inference by skipping layers to create an efficient draft model, yet existing methods often rely on static heuristics that ignore the dynamic computational overhead of attention in long-context scenarios. We propose KnapSpec, a training-free framework that reformulates draft model selection as a knapsack problem to maximize tokens-per-time throughput. By decoupling Attention and MLP layers and modeling their hardware-specific latencies as functions of context length, KnapSpec adaptively identifies optimal draft configurations on the fly via a parallel dynamic programming algorithm. Furthermore, we provide the first rigorous theoretical analysis establishing cosine similarity between hidden states as a mathematically sound proxy for the token acceptance rate. This foundation allows our method to maintain high drafting faithfulness while navigating the shifting bottlenecks of real-world hardware. Our experiments on Qwen3 and Llama3 demonstrate that KnapSpec consistently outperforms state-of-the-art SSD baselines, achieving up to 1.47x wall-clock speedup across various benchmarks. Our plug-and-play approach ensures high-speed inference for long sequences without requiring additional training or compromising the target model's output distribution.

cs.LG↗
arXiv2026

PPA-Plan: Proactive Pitfall Avoidance for Reliable Planning in Long-Context LLM Reasoning

Large language models (LLMs) struggle with reasoning over long contexts where relevant information is sparsely distributed. Although plan-and-execute frameworks mitigate this by decomposing tasks into planning and execution, their effectiveness is often limited by unreliable plan generation due to dependence on surface-level cues. Consequently, plans may be based on incorrect assumptions, and once a plan is formed, identifying what went wrong and revising it reliably becomes difficult, limiting the effectiveness of reactive refinement. To address this limitation, we propose PPA-Plan, a proactive planning strategy for long-context reasoning that focuses on preventing such failures before plan generation. PPA-Plan identifies potential logical pitfalls and false assumptions, formulates them as negative constraints, and conditions plan generation on explicitly avoiding these constraints. Experiments on long-context QA benchmarks show that executing plans generated by PPA-Plan consistently outperforms existing plan-and-execute methods and direct prompting.

cs.CL↗
arXiv2025

AcuRank: Uncertainty-Aware Adaptive Computation for Listwise Reranking

Listwise reranking with large language models (LLMs) enhances top-ranked results in retrieval-based applications. Due to the limit in context size and high inference cost of long context, reranking is typically performed over a fixed size of small subsets, with the final ranking aggregated from these partial results. This fixed computation disregards query difficulty and document distribution, leading to inefficiencies. We propose AcuRank, an adaptive reranking framework that dynamically adjusts both the amount and target of computation based on uncertainty estimates over document relevance. Using a Bayesian TrueSkill model, we iteratively refine relevance estimates until reaching sufficient confidence levels, and our explicit modeling of ranking uncertainty enables principled control over reranking behavior and avoids unnecessary updates to confident predictions. Results on the TREC-DL and BEIR benchmarks show that our method consistently achieves a superior accuracy-efficiency trade-off and scales better with compute than fixed-computation baselines. These results highlight the effectiveness and generalizability of our method across diverse retrieval tasks and LLM-based reranking models.

cs.IR↗
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↗
arXiv2025

MacRAG: Compress, Slice, and Scale-up for Multi-Scale Adaptive Context RAG

Long-context large language models (LC LLMs) combined with retrieval-augmented generation (RAG) hold strong potential for complex multi-hop and large-document tasks. However, existing RAG systems often suffer from imprecise retrieval, incomplete context coverage under constrained windows, and fragmented information from suboptimal context construction. We introduce Multi-scale Adaptive Context RAG (MacRAG), a hierarchical RAG framework that compresses and partitions documents into coarse-to-fine granularities, then adaptively merges relevant contexts through real-time chunk- and document-level expansions. By initiating with finest-level retrieval and progressively incorporating broader, higher-level context, MacRAG constructs effective query-specific long contexts, optimizing both precision and coverage. Evaluations on challenging LongBench expansions of HotpotQA, 2WikiMultihopQA, and Musique confirm MacRAG consistently surpasses baseline RAG pipelines in single- and multi-step generation using Llama-3.1-8B, Gemini-1.5-pro, and GPT-4o. Our results establish MacRAG as an efficient, scalable solution for real-world long-context, multi-hop reasoning. Our code is available at https://github.com/Leezekun/MacRAG.

cs.CL↗
arXiv2024

Detecting Training Data of Large Language Models via Expectation Maximization

Membership inference attacks (MIAs) aim to determine whether a specific example was used to train a given language model. While prior work has explored prompt-based attacks such as ReCALL, these methods rely heavily on the assumption that using known non-members as prompts reliably suppresses the model's responses to non-member queries. We propose EM-MIA, a new membership inference approach that iteratively refines prefix effectiveness and membership scores using an expectation-maximization strategy without requiring labeled non-member examples. To support controlled evaluation, we introduce OLMoMIA, a benchmark that enables analysis of MIA robustness under systematically varied distributional overlap and difficulty. Experiments on WikiMIA and OLMoMIA show that EM-MIA outperforms existing baselines, particularly in settings with clear distributional separability. We highlight scenarios where EM-MIA succeeds in practical settings with partial distributional overlap, while failure cases expose fundamental limitations of current MIA methods under near-identical conditions. We release our code and evaluation pipeline to encourage reproducible and robust MIA research.

cs.CL↗
arXiv2023

Towards standardizing Korean Grammatical Error Correction: Datasets and Annotation

Research on Korean grammatical error correction (GEC) is limited, compared to other major languages such as English. We attribute this problematic circumstance to the lack of a carefully designed evaluation benchmark for Korean GEC. In this work, we collect three datasets from different sources (Kor-Lang8, Kor-Native, and Kor-Learner) that covers a wide range of Korean grammatical errors. Considering the nature of Korean grammar, We then define 14 error types for Korean and provide KAGAS (Korean Automatic Grammatical error Annotation System), which can automatically annotate error types from parallel corpora. We use KAGAS on our datasets to make an evaluation benchmark for Korean, and present baseline models trained from our datasets. We show that the model trained with our datasets significantly outperforms the currently used statistical Korean GEC system (Hanspell) on a wider range of error types, demonstrating the diversity and usefulness of the datasets. The implementations and datasets are open-sourced.

cs.CL↗
arXiv2022

Bridging the Training-Inference Gap for Dense Phrase Retrieval

Building dense retrievers requires a series of standard procedures, including training and validating neural models and creating indexes for efficient search. However, these procedures are often misaligned in that training objectives do not exactly reflect the retrieval scenario at inference time. In this paper, we explore how the gap between training and inference in dense retrieval can be reduced, focusing on dense phrase retrieval (Lee et al., 2021) where billions of representations are indexed at inference. Since validating every dense retriever with a large-scale index is practically infeasible, we propose an efficient way of validating dense retrievers using a small subset of the entire corpus. This allows us to validate various training strategies including unifying contrastive loss terms and using hard negatives for phrase retrieval, which largely reduces the training-inference discrepancy. As a result, we improve top-1 phrase retrieval accuracy by 2~3 points and top-20 passage retrieval accuracy by 2~4 points for open-domain question answering. Our work urges modeling dense retrievers with careful consideration of training and inference via efficient validation while advancing phrase retrieval as a general solution for dense retrieval.

cs.CL↗
arXiv2022

Consistency Training with Virtual Adversarial Discrete Perturbation

Consistency training regularizes a model by enforcing predictions of original and perturbed inputs to be similar. Previous studies have proposed various augmentation methods for the perturbation but are limited in that they are agnostic to the training model. Thus, the perturbed samples may not aid in regularization due to their ease of classification from the model. In this context, we propose an augmentation method of adding a discrete noise that would incur the highest divergence between predictions. This virtual adversarial discrete noise obtained by replacing a small portion of tokens while keeping original semantics as much as possible efficiently pushes a training model's decision boundary. Experimental results show that our proposed method outperforms other consistency training baselines with text editing, paraphrasing, or a continuous noise on semi-supervised text classification tasks and a robustness benchmark

cs.CL↗
arXiv2021

Dynamic-TinyBERT: Boost TinyBERT's Inference Efficiency by Dynamic Sequence Length

Limited computational budgets often prevent transformers from being used in production and from having their high accuracy utilized. TinyBERT addresses the computational efficiency by self-distilling BERT into a smaller transformer representation having fewer layers and smaller internal embedding. However, TinyBERT's performance drops when we reduce the number of layers by 50%, and drops even more abruptly when we reduce the number of layers by 75% for advanced NLP tasks such as span question answering. Additionally, a separate model must be trained for each inference scenario with its distinct computational budget. In this work we present Dynamic-TinyBERT, a TinyBERT model that utilizes sequence-length reduction and Hyperparameter Optimization for enhanced inference efficiency per any computational budget. Dynamic-TinyBERT is trained only once, performing on-par with BERT and achieving an accuracy-speedup trade-off superior to any other efficient approaches (up to 3.3x with <1% loss-drop). Upon publication, the code to reproduce our work will be open-sourced.

cs.CL↗
arXiv2021

SSMix: Saliency-Based Span Mixup for Text Classification

Data augmentation with mixup has shown to be effective on various computer vision tasks. Despite its great success, there has been a hurdle to apply mixup to NLP tasks since text consists of discrete tokens with variable length. In this work, we propose SSMix, a novel mixup method where the operation is performed on input text rather than on hidden vectors like previous approaches. SSMix synthesizes a sentence while preserving the locality of two original texts by span-based mixing and keeping more tokens related to the prediction relying on saliency information. With extensive experiments, we empirically validate that our method outperforms hidden-level mixup methods on a wide range of text classification benchmarks, including textual entailment, sentiment classification, and question-type classification. Our code is available at https://github.com/clovaai/ssmix.

cs.CL↗
arXiv2021

Length-Adaptive Transformer: Train Once with Length Drop, Use Anytime with Search

Despite transformers' impressive accuracy, their computational cost is often prohibitive to use with limited computational resources. Most previous approaches to improve inference efficiency require a separate model for each possible computational budget. In this paper, we extend PoWER-BERT (Goyal et al., 2020) and propose Length-Adaptive Transformer that can be used for various inference scenarios after one-shot training. We train a transformer with LengthDrop, a structural variant of dropout, which stochastically determines a sequence length at each layer. We then conduct a multi-objective evolutionary search to find a length configuration that maximizes the accuracy and minimizes the efficiency metric under any given computational budget. Additionally, we significantly extend the applicability of PoWER-BERT beyond sequence-level classification into token-level classification with Drop-and-Restore process that drops word-vectors temporarily in intermediate layers and restores at the last layer if necessary. We empirically verify the utility of the proposed approach by demonstrating the superior accuracy-efficiency trade-off under various setups, including span-based question answering and text classification. Code is available at https://github.com/clovaai/length-adaptive-transformer.

cs.CL↗
arXiv2021

Two-stage Textual Knowledge Distillation for End-to-End Spoken Language Understanding

End-to-end approaches open a new way for more accurate and efficient spoken language understanding (SLU) systems by alleviating the drawbacks of traditional pipeline systems. Previous works exploit textual information for an SLU model via pre-training with automatic speech recognition or fine-tuning with knowledge distillation. To utilize textual information more effectively, this work proposes a two-stage textual knowledge distillation method that matches utterance-level representations and predicted logits of two modalities during pre-training and fine-tuning, sequentially. We use vq-wav2vec BERT as a speech encoder because it captures general and rich features. Furthermore, we improve the performance, especially in a low-resource scenario, with data augmentation methods by randomly masking spans of discrete audio tokens and contextualized hidden representations. Consequently, we push the state-of-the-art on the Fluent Speech Commands, achieving 99.7% test accuracy in the full dataset setting and 99.5% in the 10% subset setting. Throughout the ablation studies, we empirically verify that all used methods are crucial to the final performance, providing the best practice for spoken language understanding. Code is available at https://github.com/clovaai/textual-kd-slu.

cs.CL↗
arXiv2021

ST-BERT: Cross-modal Language Model Pre-training For End-to-end Spoken Language Understanding

Language model pre-training has shown promising results in various downstream tasks. In this context, we introduce a cross-modal pre-trained language model, called Speech-Text BERT (ST-BERT), to tackle end-to-end spoken language understanding (E2E SLU) tasks. Taking phoneme posterior and subword-level text as an input, ST-BERT learns a contextualized cross-modal alignment via our two proposed pre-training tasks: Cross-modal Masked Language Modeling (CM-MLM) and Cross-modal Conditioned Language Modeling (CM-CLM). Experimental results on three benchmarks present that our approach is effective for various SLU datasets and shows a surprisingly marginal performance degradation even when 1% of the training data are available. Also, our method shows further SLU performance gain via domain-adaptive pre-training with domain-specific speech-text pair data.

cs.CL↗
arXiv2021

AdamP: Slowing Down the Slowdown for Momentum Optimizers on Scale-invariant Weights

Normalization techniques are a boon for modern deep learning. They let weights converge more quickly with often better generalization performances. It has been argued that the normalization-induced scale invariance among the weights provides an advantageous ground for gradient descent (GD) optimizers: the effective step sizes are automatically reduced over time, stabilizing the overall training procedure. It is often overlooked, however, that the additional introduction of momentum in GD optimizers results in a far more rapid reduction in effective step sizes for scale-invariant weights, a phenomenon that has not yet been studied and may have caused unwanted side effects in the current practice. This is a crucial issue because arguably the vast majority of modern deep neural networks consist of (1) momentum-based GD (e.g. SGD or Adam) and (2) scale-invariant parameters. In this paper, we verify that the widely-adopted combination of the two ingredients lead to the premature decay of effective step sizes and sub-optimal model performances. We propose a simple and effective remedy, SGDP and AdamP: get rid of the radial component, or the norm-increasing direction, at each optimizer step. Because of the scale invariance, this modification only alters the effective step sizes without changing the effective update directions, thus enjoying the original convergence properties of GD optimizers. Given the ubiquity of momentum GD and scale invariance in machine learning, we have evaluated our methods against the baselines on 13 benchmarks. They range from vision tasks like classification (e.g. ImageNet), retrieval (e.g. CUB and SOP), and detection (e.g. COCO) to language modelling (e.g. WikiText) and audio classification (e.g. DCASE) tasks. We verify that our solution brings about uniform gains in those benchmarks. Source code is available at https://github.com/clovaai/AdamP.

cs.LG↗
arXiv2020

Large Product Key Memory for Pretrained Language Models

Product key memory (PKM) proposed by Lample et al. (2019) enables to improve prediction accuracy by increasing model capacity efficiently with insignificant computational overhead. However, their empirical application is only limited to causal language modeling. Motivated by the recent success of pretrained language models (PLMs), we investigate how to incorporate large PKM into PLMs that can be finetuned for a wide variety of downstream NLP tasks. We define a new memory usage metric, and careful observation using this metric reveals that most memory slots remain outdated during the training of PKM-augmented models. To train better PLMs by tackling this issue, we propose simple but effective solutions: (1) initialization from the model weights pretrained without memory and (2) augmenting PKM by addition rather than replacing a feed-forward network. We verify that both of them are crucial for the pretraining of PKM-augmented PLMs, enhancing memory utilization and downstream performance. Code and pretrained weights are available at https://github.com/clovaai/pkm-transformers.

cs.CL↗