arXiv subjects

Yukyung Lee

Publications and source records attributed to Yukyung Lee.

18 recordsLinked to original sources

arXiv2026

Harbor Adapters and Harbor-Index: Infrastructure and a Curated Meta-Dataset for Large-Scale Agentic Evaluation

Evaluating agents on the growing number of agentic benchmarks is challenging because they often require complex environments and agent integrations. We introduce Harbor Adapters, a unified evaluation infrastructure for agentic benchmarks. Our work makes three contributions. First, we develop benchmark adapters that port more than 80 benchmarks to evaluate arbitrary agents, and validate them through rigorous code review and parity experiments. Second, we conduct a large-scale evaluation of 8 models spanning capability tiers across 54 benchmarks; every model is run with Terminus-2 and with one of 3 native harnesses. This enables a broader analysis of agent capabilities and failure modes than was previously possible. Third, we introduce Harbor-Index, a curated set of 82 difficult, diverse, and high-quality tasks spanning 29 benchmarks, refined from the adapted suite through difficulty filtering, AI and human audit, and an audit-and-fix loop. Harbor-Index preserves the challenge and breadth of large-scale agentic evaluations while being affordable to run; no evaluated model-harness configuration exceeds 30% pass rate, and the strongest (GPT-5.5 with Codex) reaches 28.0%. We release the adapters, evaluation results, in-depth analysis, and Harbor-Index as open-source artifacts to support more reliable and comprehensive evaluation of language-model agents.

cs.AI↗

arXiv2026

PGMem: Tightly Coupled Persona-Memory Graph for Lifelong Personalized Agents

Long-term personalized dialogue agents must track user preferences as their personas evolve. Existing memory systems organize past events well, but store personas as flat profiles detached from the events that justify them. This loose coupling leads to the memory-persona validity gap and the persona-aware retrieval gap. We propose PGMem, a heterogeneous persona-memory graph that connects event and persona nodes through typed provenance and evidence edges, keeping each persona signal traceable to the events that support or revise it. At retrieval time, PGMem expands from query-relevant seeds and ranks signals by evidential validity. Across three benchmarks with small language model backbones, PGMem consistently outperforms summary-based, persona-aware, graph-structured, and agentic memory baselines, and improves performance as the context grows. The source code of PGMem is available at https://github.com/wonjunchoi23/pgmem/

cs.CL↗

arXiv2026

LLMs Learn Better In-Context from Rules than from Examples

Large language models (LLMs) exhibit in-context learning capabilities, where they can learn new tasks from prompt contexts without weight updates. We compare the learning efficacies of two prominent modes of in-context learning: (1) learning from descriptions of rules (instruction following); and (2) learning from examples of input-output demonstrations (few-shot prompting). Through five learning tasks that cover diverse domains (games, arithmetic, linguistic inferences), we compare two modes of learning (rules vs. examples) specifying the same underlying task. We furthermore explore model and task properties that modulate the learning efficacies. We find that models generally learn more reliably from rules than from examples alone, and additional examples on top of rules or simply scaling up the number of examples do not lead to consistent and significant gains. Instruction tuning amplifies the benefit of rule-based learning while keeping example-based learning capacities intact. Surprisingly, we find no privileged effect of example-based learning in base models, and rules still lead to gains in algebraic task domains. Overall, the comparative efficacy of rules over examples is larger when the task recruits algebraic abstractions and computations, and smaller when the task requires distributional sensitivity and/or recruits parametric knowledge.

cs.CL↗

arXiv2026

AI Writers Have a Consistent Stylometric Footprint, but AI Editors Do Not

Text generated by large language models (LLMs) has been shown to be stylometrically distinct from human-written text \citep{andreDetectingAIAuthorship2023, shahDetectingUnmaskingAIGenerated2023, oparaStyloAIDistinguishingAIGenerated2024, soto2024fewshot, liLinguisticDifferencesAI2025, selviogluFeatureExtractionAnalysis2025}. But LLMs are increasingly used not only to generate text but also to edit human writing, and it is unclear whether the two leave the same trace. We show that AI generation leaves a consistent ``stylometric footprint'': a small subset of features, primarily entropy and lexical diversity, consistently separates AI-generated text from human writing across 8 LLMs and 5 domains, while the remaining features depend heavily on the domain and generator. AI editing, however, does not reproduce the same footprint. Relative to their human-written sources, AI-edited texts show only a small increase in lexical diversity and a decrease in entropy, rather than the joint increase that characterizes AI generation. Lexical density, which contributes little to generation, instead becomes the dominant editing-associated signal. Stylometric features therefore separate AI-edited text from AI-generated text but are substantially less effective at separating it from human-written text. Our results suggest that ``AI text'' is not a single phenomenon: generation and editing leave qualitatively different stylometric traces and should be studied separately.

cs.CL↗

arXiv2026

Cliff Tokens: Identifying Single-Token Failure Triggers in LLM Mathematical Reasoning

Large language models (LLMs) reach high accuracy in mathematical reasoning, but individual traces on the same problem diverge; some arrive at the correct answer while others fail. Prior work analyzes failure at the step, chunk, or sentence level, or at tokens where failure has already occurred. Neither identifies the precise token that triggers the shift toward failure. We introduce the cliff token, a token where the token-wise potential drops significantly under an adaptive threshold that scales with the local token-wise potential, based on a one-sided two-proportion z-test. Across seven models and three mathematical reasoning benchmarks (GSM1K, MATH500, AIME 2025), cliff tokens act as failure triggers; deleting the first cliff token and resampling recovers pass@64 to 1.0, while keeping it limits recovery to between 0.71 and 1.00. We further introduce a cliff taxonomy of deterministic, uncertain, and sampled-off cliffs, defined by greedy choice and token entropy. Each type has distinct probabilistic characteristics, and the taxonomy generalizes across model scales. Finally, we validate the taxonomy via single-token preference optimization at cliff positions (Cliff-DPO). Trained on GSM8K, Cliff-DPO improves accuracy across benchmarks by up to +6.6. Optimizing at uncertain and sampled-off cliffs improves reasoning, while deterministic cliffs do not.

cs.AI↗

arXiv2026

Can Structural Cues Save LLMs? Evaluating Language Models in Massive Document Streams

Evaluating language models in streaming environments is critical, yet underexplored. Existing benchmarks either focus on single complex events or provide curated inputs for each query, and do not evaluate models under the conflicts that arise when multiple concurrent events are mixed within the same document stream. We introduce StreamBench, a benchmark built from major news stories in 2016 and 2025, comprising 605 events and 15,354 documents across three tasks: Topic Clustering, Temporal Question Answering, and Summarization. To diagnose how models fail, we compare performance with and without structural cues, which organize key facts by event. We find that structural cues improve performance on clustering (up to +4.37%) and temporal QA (up to +9.63%), helping models locate relevant information and separate distinct events. While temporal reasoning remains an open challenge inherent to current LLMs, consistent gains across tasks show that structural cues are a promising direction for future work in massive document streams.

cs.CL↗

arXiv2026

CIRF: Tokenizing Chain-of-Thoughts into Reusable Functional Units for Efficient Latent Reasoning in Large Language Models

Implicit Chain-of-Thought (CoT) reduces the inference cost of large language models by internalizing the explicit rationales. However, existing approaches typically lack alignment with explicit rationales and adaptivity to example complexity. In this work, we propose CIRF (\textit{\underline{C}hain-of-thoughts \underline{I}nto \underline{R}eusable \underline{F}unctional units}), an implicit CoT framework that performs reasoning as a dynamic sequence of discrete functional tokens. CIRF assigns a functional token to each semantically coherent reasoning unit in explicit CoT traces. The model is then fine-tuned to autoregressively generate functional tokens and their optional results, followed by the final answer. This design aligns latent reasoning with a sequence of functional units, facilitating parallel training, explicit rationale alignment, and adaptive reasoning. Extensive experiments on mathematical, symbolic, and commonsense reasoning benchmarks show that CIRF provides a favorable accuracy-latency trade-off compared with state-of-the-art implicit CoT methods. Further analyses demonstrate that CIRF constructs distinct, interpretable functional tokens, leading to consistent performance improvements.

cs.CL↗

arXiv2026

RExBench: Can coding agents autonomously implement AI research extensions?

Agents based on Large Language Models (LLMs) have shown promise for performing sophisticated software engineering tasks autonomously. In addition, there has been progress towards developing agents that can perform parts of the research pipeline in machine learning and the natural sciences. We argue that research extension and its implementation is a critical capability for such systems, and introduce RExBench to support the evaluation of this capability. RExBench is a benchmark consisting of realistic extensions of 12 research papers that aim to investigate novel research hypotheses. Each task is set up as an extension to an existing research paper and codebase, accompanied by domain expert-written instructions. RExBench is robust to data contamination and supports an automatic evaluation infrastructure that executes agent outputs to determine whether the success criteria are met. We use this benchmark to evaluate 12 LLM agents implemented using two different frameworks, aider and OpenHands. We find that all agents fail to autonomously implement the majority of the extensions, with the best agent achieving around a 33% success rate. Although the success rate improves with additional human-written hints, the best performance under this setting remains below 44%. This indicates that current agents are still short of being able to handle realistic research extension tasks without substantial human guidance.

cs.CL↗

arXiv2025

CheckEval: A reliable LLM-as-a-Judge framework for evaluating text generation using checklists

Existing LLM-as-a-Judge approaches for evaluating text generation suffer from rating inconsistencies, with low agreement and high rating variance across different evaluator models. We attribute this to subjective evaluation criteria combined with Likert scale scoring in existing protocols. To address this issue, we introduce CheckEval, a checklist-based evaluation framework that improves rating reliability via decomposed binary questions. Through experiments with 12 evaluator models across multiple datasets, we first demonstrate that CheckEval strongly correlates with human judgments. More importantly, CheckEval dramatically improves the average agreement across evaluator models by 0.45 and reduces the score variance. CheckEval scores furthermore have the benefit of being more interpretable because it decomposes evaluation criteria into traceable binary decisions, allowing analyses of specific attributes driving quality judgments.

cs.CL↗

arXiv2025

Navigating the Path of Writing: Outline-guided Text Generation with Large Language Models

Large Language Models (LLMs) have impacted the writing process, enhancing productivity by collaborating with humans in content creation platforms. However, generating high-quality, user-aligned text to satisfy real-world content creation needs remains challenging. We propose WritingPath, a framework that uses explicit outlines to guide LLMs in generating goal-oriented, high-quality text. Our approach draws inspiration from structured writing planning and reasoning paths, focusing on reflecting user intentions throughout the writing process. To validate our approach in real-world scenarios, we construct a diverse dataset from unstructured blog posts to benchmark writing performance and introduce a comprehensive evaluation framework assessing the quality of outlines and generated texts. Our evaluations with various LLMs demonstrate that the WritingPath approach significantly enhances text quality according to evaluations by both LLMs and professional writers.

cs.CL↗

arXiv2024

A Gradient Accumulation Method for Dense Retriever under Memory Constraint

InfoNCE loss is commonly used to train dense retriever in information retrieval tasks. It is well known that a large batch is essential to stable and effective training with InfoNCE loss, which requires significant hardware resources. Due to the dependency of large batch, dense retriever has bottleneck of application and research. Recently, memory reduction methods have been broadly adopted to resolve the hardware bottleneck by decomposing forward and backward or using a memory bank. However, current methods still suffer from slow and unstable training. To address these issues, we propose Contrastive Accumulation (ContAccum), a stable and efficient memory reduction method for dense retriever trains that uses a dual memory bank structure to leverage previously generated query and passage representations. Experiments on widely used five information retrieval datasets indicate that ContAccum can surpass not only existing memory reduction methods but also high-resource scenario. Moreover, theoretical analysis and experimental results confirm that ContAccum provides more stable dual-encoder training than current memory bank utilization methods.

cs.IR↗

arXiv2023

RAPID: Training-free Retrieval-based Log Anomaly Detection with PLM considering Token-level information

As the IT industry advances, system log data becomes increasingly crucial. Many computer systems rely on log texts for management due to restricted access to source code. The need for log anomaly detection is growing, especially in real-world applications, but identifying anomalies in rapidly accumulating logs remains a challenging task. Traditional deep learning-based anomaly detection models require dataset-specific training, leading to corresponding delays. Notably, most methods only focus on sequence-level log information, which makes the detection of subtle anomalies harder, and often involve inference processes that are difficult to utilize in real-time. We introduce RAPID, a model that capitalizes on the inherent features of log data to enable anomaly detection without training delays, ensuring real-time capability. RAPID treats logs as natural language, extracting representations using pre-trained language models. Given that logs can be categorized based on system context, we implement a retrieval-based technique to contrast test logs with the most similar normal logs. This strategy not only obviates the need for log-specific training but also adeptly incorporates token-level information, ensuring refined and robust detection, particularly for unseen logs. We also propose the core set technique, which can reduce the computational cost needed for comparison. Experimental results show that even without training on log data, RAPID demonstrates competitive performance compared to prior models and achieves the best performance on certain datasets. Through various research questions, we verified its capability for real-time detection without delay.

cs.LG↗

arXiv2023

LAnoBERT: System Log Anomaly Detection based on BERT Masked Language Model

The system log generated in a computer system refers to large-scale data that are collected simultaneously and used as the basic data for determining errors, intrusion and abnormal behaviors. The aim of system log anomaly detection is to promptly identify anomalies while minimizing human intervention, which is a critical problem in the industry. Previous studies performed anomaly detection through algorithms after converting various forms of log data into a standardized template using a parser. Particularly, a template corresponding to a specific event should be defined in advance for all the log data using which the information within the log key may get lost. In this study, we propose LAnoBERT, a parser free system log anomaly detection method that uses the BERT model, exhibiting excellent natural language processing performance. The proposed method, LAnoBERT, learns the model through masked language modeling, which is a BERT-based pre-training method, and proceeds with unsupervised learning-based anomaly detection using the masked language modeling loss function per log key during the test process. In addition, we also propose an efficient inference process to establish a practically applicable pipeline to the actual system. Experiments on three well-known log datasets, i.e., HDFS, BGL, and Thunderbird, show that not only did LAnoBERT yield a higher anomaly detection performance compared to unsupervised learning-based benchmark models, but also it resulted in a comparable performance with supervised learning-based benchmark models.

cs.LG↗

arXiv2023

DSTEA: Improving Dialogue State Tracking via Entity Adaptive Pre-training

Dialogue State Tracking (DST) is critical for comprehensively interpreting user and system utterances, thereby forming the cornerstone of efficient dialogue systems. Despite past research efforts focused on enhancing DST performance through alterations to the model structure or integrating additional features like graph relations, they often require additional pre-training with external dialogue corpora. In this study, we propose DSTEA, improving Dialogue State Tracking via Entity Adaptive pre-training, which can enhance the encoder through by intensively training key entities in dialogue utterances. DSTEA identifies these pivotal entities from input dialogues utilizing four different methods: ontology information, named-entity recognition, the spaCy, and the flair library. Subsequently, it employs selective knowledge masking to train the model effectively. Remarkably, DSTEA only requires pre-training without the direct infusion of extra knowledge into the DST model. This approach resulted in substantial performance improvements of four robust DST models on MultiWOZ 2.0, 2.1, and 2.2, with joint goal accuracy witnessing an increase of up to 2.69% (from 52.41% to 55.10%). Further validation of DSTEA's efficacy was provided through comparative experiments considering various entity types and different entity adaptive pre-training configurations such as masking strategy and masking rate.

cs.CL↗

arXiv2023

Painsight: An Extendable Opinion Mining Framework for Detecting Pain Points Based on Online Customer Reviews

As the e-commerce market continues to expand and online transactions proliferate, customer reviews have emerged as a critical element in shaping the purchasing decisions of prospective buyers. Previous studies have endeavored to identify key aspects of customer reviews through the development of sentiment analysis models and topic models. However, extracting specific dissatisfaction factors remains a challenging task. In this study, we delineate the pain point detection problem and propose Painsight, an unsupervised framework for automatically extracting distinct dissatisfaction factors from customer reviews without relying on ground truth labels. Painsight employs pre-trained language models to construct sentiment analysis and topic models, leveraging attribution scores derived from model gradients to extract dissatisfaction factors. Upon application of the proposed methodology to customer review data spanning five product categories, we successfully identified and categorized dissatisfaction factors within each group, as well as isolated factors for each type. Notably, Painsight outperformed benchmark methods, achieving substantial performance enhancements and exceptional results in human evaluations.

cs.AI↗

arXiv2022

Oh My Mistake!: Toward Realistic Dialogue State Tracking including Turnback Utterances

The primary purpose of dialogue state tracking (DST), a critical component of an end-to-end conversational system, is to build a model that responds well to real-world situations. Although we often change our minds from time to time during ordinary conversations, current benchmark datasets do not adequately reflect such occurrences and instead consist of over-simplified conversations, in which no one changes their mind during a conversation. As the main question inspiring the present study, "Are current benchmark datasets sufficiently diverse to handle casual conversations in which one changes their mind after a certain topic is over?" We found that the answer is "No" because DST models cannot refer to previous user preferences when template-based turnback utterances are injected into the dataset. Even in the the simplest mind-changing (turnback) scenario, the performance of DST models significantly degenerated. However, we found that this performance degeneration can be recovered when the turnback scenarios are explicitly designed in the training set, implying that the problem is not with the DST models but rather with the construction of the benchmark dataset.

cs.CL↗

arXiv2022

Mismatch between Multi-turn Dialogue and its Evaluation Metric in Dialogue State Tracking

Dialogue state tracking (DST) aims to extract essential information from multi-turn dialogue situations and take appropriate actions. A belief state, one of the core pieces of information, refers to the subject and its specific content, and appears in the form of domain-slot-value. The trained model predicts "accumulated" belief states in every turn, and joint goal accuracy and slot accuracy are mainly used to evaluate the prediction; however, we specify that the current evaluation metrics have a critical limitation when evaluating belief states accumulated as the dialogue proceeds, especially in the most used MultiWOZ dataset. Additionally, we propose relative slot accuracy to complement existing metrics. Relative slot accuracy does not depend on the number of predefined slots, and allows intuitive evaluation by assigning relative scores according to the turn of each dialogue. This study also encourages not solely the reporting of joint goal accuracy, but also various complementary metrics in DST tasks for the sake of a realistic evaluation.

cs.CL↗

arXiv2020

Multi$^2$OIE: Multilingual Open Information Extraction Based on Multi-Head Attention with BERT

In this paper, we propose Multi$^2$OIE, which performs open information extraction (open IE) by combining BERT with multi-head attention. Our model is a sequence-labeling system with an efficient and effective argument extraction method. We use a query, key, and value setting inspired by the Multimodal Transformer to replace the previously used bidirectional long short-term memory architecture with multi-head attention. Multi$^2$OIE outperforms existing sequence-labeling systems with high computational efficiency on two benchmark evaluation datasets, Re-OIE2016 and CaRB. Additionally, we apply the proposed method to multilingual open IE using multilingual BERT. Experimental results on new benchmark datasets introduced for two languages (Spanish and Portuguese) demonstrate that our model outperforms other multilingual systems without training data for the target languages.

cs.CL↗