Donghun Lee

Eugene Vorontsov, Alican Bozkurt, Adam Casson et al.

The use of artificial intelligence to enable precision medicine and decision support systems through the analysis of pathology images has the potential to revolutionize the diagnosis and treatment of cancer. Such applications will depend on models' abilities to capture the diverse patterns observed in pathology images. To address this challenge, we present Virchow, a foundation model for computational pathology. Using self-supervised learning empowered by the DINOv2 algorithm, Virchow is a vision transformer model with 632 million parameters trained on 1.5 million hematoxylin and eosin stained whole slide images from diverse tissue and specimen types, which is orders of magnitude more data than previous works. The Virchow model enables the development of a pan-cancer detection system with 0.949 overall specimen-level AUC across 17 different cancer types, while also achieving 0.937 AUC on 7 rare cancer types. The Virchow model sets the state-of-the-art on the internal and external image tile level benchmarks and slide level biomarker prediction tasks. The gains in performance highlight the importance of training on massive pathology image datasets, suggesting scaling up the data and network architecture can improve the accuracy for many high-impact computational pathology applications where limited amounts of training data are available.

Joosung Lee, Minsik Oh, Donghun Lee

The use of persona-grounded retrieval-based chatbots is crucial for personalized conversations, but there are several challenges that need to be addressed. 1) In general, collecting persona-grounded corpus is very expensive. 2) The chatbot system does not always respond in consideration of persona at real applications. To address these challenges, we propose a plug-and-play persona prompting method. Our system can function as a standard open-domain chatbot if persona information is not available. We demonstrate that this approach performs well in the zero-shot setting, which reduces the dependence on persona-ground training data. This makes it easier to expand the system to other languages without the need to build a persona-grounded corpus. Additionally, our model can be fine-tuned for even better performance. In our experiments, the zero-shot model improved the standard model by 7.71 and 1.04 points in the original persona and revised persona, respectively. The fine-tuned model improved the previous state-of-the-art system by 1.95 and 3.39 points in the original persona and revised persona, respectively. To the best of our knowledge, this is the first attempt to solve the problem of personalized response selection using prompt sequences. Our code is available on github~\footnote{https://github.com/rungjoo/plug-and-play-prompt-persona}.

Jaehyuk Lee, Hanyoung Kim, Yanggee Kim et al.

Vision Transformers (ViTs) face severe computational bottlenecks due to the quadratic complexity of self-attention at high resolutions. Existing token reduction methods rely on local metrics - such as single-layer attention scores - that are inherently vulnerable to the attention sink phenomenon, where uninformative tokens are paradoxically preserved over salient foreground objects. We propose ASAP (Attention Sink Anchored Pruning), a training-free framework that recasts this sink as a feature. Modeling ViT information flow as a Lazy Random Walk, ASAP identifies the sink as a dominant accumulator of probability mass. By computing the diffusion distance to the sink within the cumulative transition matrix, ASAP partitions tokens via Radial Diffusion Clustering and compresses background redundancy through Transition Weight Pooling in a single shot. Extensive experiments across image, video, and vision-language tasks demonstrate ASAP outperforms state-of-the-art methods, accelerating throughput by up to 48% while maintaining - or even exceeding - baseline accuracy.

Yejin Jeong, Donghun Lee

Knee osteoarthritis (KOA) is a universal chronic musculoskeletal disorders worldwide, making early diagnosis crucial. Currently, the Kellgren and Lawrence (KL) grading system is widely used to assess KOA severity. However, its high inter-observer variability and subjectivity hinder diagnostic consistency. To address these limitations, automated diagnostic techniques using deep learning have been actively explored in recent years. In this study, we propose a CLIP-based framework (CLIP-KOA) to enhance the consistency and reliability of KOA grade prediction. To achieve this, we introduce a learning approach that integrates image and text information and incorporate Symmetry Loss and Consistency Loss to ensure prediction consistency between the original and flipped images. CLIP-KOA achieves state-of-the-art accuracy of 71.86\% on KOA severity prediction task, and ablation studies show that CLIP-KOA has 2.36\% improvement in accuracy over the standard CLIP model due to our contribution. This study shows a novel direction for data-driven medical prediction not only to improve reliability of fine-grained diagnosis and but also to explore multimodal methods for medical image analysis. Our code is available at https://github.com/anonymized-link.

Taehun Cha, Daniel Beaglehole, Adityanarayanan Radhakrishnan et al.

Understanding how deep neural networks learn representations remains a central challenge in machine learning theory. In this work, we propose a feature-centric framework for analyzing neural network training by relating weight updates to feature evolution. We introduce a simple identity, the Feature Learning Equation, which identifies the weight Gram matrix as the key object capturing feature dynamics. This enables us to interpret gradient descent as implicitly inducing a hypothetical evolution of features, whose covariance structure - termed the Virtual Covariance - characterizes how representations evolve during training. Building on this perspective, we introduce Target Linearity, a measure quantifying the linear alignment between features and targets. By analyzing the training and layer-wise dynamics, we show that deep networks learn to sequentially transform representations toward target-linear structure. This linearization perspective provides a unified interpretation of several empirical phenomena, including Neural Collapse and linear interpolation in generative models.

Taehun Cha, Donghun Lee

In this work, we show the pre-trained language models return distinguishable generation probability and uncertainty distribution to unfaithfully hallucinated texts, regardless of their size and structure. By examining 24 models on 6 data sets, we find out that 88-98% of cases return statistically significantly distinguishable generation probability and uncertainty distributions. Using this general phenomenon, we showcase a hallucination-reducing training algorithm. Our algorithm outperforms other baselines by achieving higher faithfulness metrics while maintaining sound general text quality measures.

Kanana LLM Team, Yunju Bak, Hojin Lee et al.

We introduce Kanana, a series of bilingual language models that demonstrate exceeding performance in Korean and competitive performance in English. The computational cost of Kanana is significantly lower than that of state-of-the-art models of similar size. The report details the techniques employed during pre-training to achieve compute-efficient yet competitive models, including high quality data filtering, staged pre-training, depth up-scaling, and pruning and distillation. Furthermore, the report outlines the methodologies utilized during the post-training of the Kanana models, encompassing supervised fine-tuning and preference optimization, aimed at enhancing their capability for seamless interaction with users. Lastly, the report elaborates on plausible approaches used for language model adaptation to specific scenarios, such as embedding, retrieval augmented generation, and function calling. The Kanana model series spans from 2.1B to 32.5B parameters with 2.1B models (base, instruct, embedding) publicly released to promote research on Korean language models.

Sunghee Jung, Donghun Lee, Shinbok Lee et al.

Tool-Augmented Larage Language Models (TA-LLMs) have shown promise in real-world applications, but face challenges in handling incomplete queries and out-of-scope requests. While existing approaches rely mainly on Supervised Fine-Tuning with expert trajectories, we propose DiaTool-DPO, a novel method that enhances TA-LLM's dialogue capabilities through Direct Preference Optimization. We model TA-LLM interactions as a Markov Decision Process with 5 distinct dialogue states and categorize user queries into 3 types based on their state transition trajectories. We automatically construct paired trajectory datasets of correct and incorrect dialogue flows and introduce a specialized objective loss for dialogue control. Our comprehensive evaluation demonstrates that DiaTool-DPO approaches GPT-4o's performance (94.8% in information gathering, 91% in tool call rejection) with substantial improvements over baseline (44% and 9.6% respectively) while maintaining core functionality. Our approach opens new possibilities for developing TA-LLMs that can handle diverse real-world scenarios without requiring additional expert demonstrations or human labeling.

Taehun Cha, Donghun Lee

In causal inference, randomized experiment is a de facto method to overcome various theoretical issues in observational study. However, the experimental design requires expensive costs, so an efficient experimental design is necessary. We propose ABC3, a Bayesian active learning policy for causal inference. We show a policy minimizing an estimation error on conditional average treatment effect is equivalent to minimizing an integrated posterior variance, similar to Cohn criteria \citep{cohn1994active}. We theoretically prove ABC3 also minimizes an imbalance between the treatment and control groups and the type 1 error probability. Imbalance-minimizing characteristic is especially notable as several works have emphasized the importance of achieving balance. Through extensive experiments on real-world data sets, ABC3 achieves the highest efficiency, while empirically showing the theoretical results hold.

Woon Ryong Kim, Jaeheun Jung, Jeong Un Ha et al.

Dimensional synthesis of planar four-bar mechanisms is a challenging inverse problem in kinematics, requiring the determination of mechanism dimensions from desired motion specifications. We propose a data-driven framework that bypasses traditional equation-solving and optimization by leveraging supervised learning. Our method combines a synthetic dataset, an LSTM-based neural network for handling sequential precision points, and a Mixture of Experts (MoE) architecture tailored to different linkage types. Each expert model is trained on type-specific data and guided by a type-specifying layer, enabling both single-type and multi-type synthesis. A novel simulation metric evaluates prediction quality by comparing desired and generated motions. Experiments show our approach produces accurate, defect-free linkages across various configurations. This enables intuitive and efficient mechanism design, even for non-expert users, and opens new possibilities for scalable and flexible synthesis in kinematic design.

Jaeheun Jung, Bosung Jung, Suhyun Bae et al.

Machine unlearning seeks to remove the influence of particular data or class from trained models to meet privacy, legal, or ethical requirements. Existing unlearning methods tend to forget shallowly: phenomenon of an unlearned model pretend to forget by adjusting only the model response, while its internal representations retain information sufficiently to restore the forgotten data or behavior. We empirically confirm the widespread shallowness by reverting the forgetting effect of various unlearning methods via training-free performance recovery attack and gradient-inversion-based data reconstruction attack. To address this vulnerability fundamentally, we define a theoretical criterion of ``deep forgetting'' based on one-point-contraction of feature representations of data to forget. We also propose an efficient approximation algorithm, and use it to construct a novel general-purpose unlearning algorithm: One-Point-Contraction (OPC). Empirical evaluations on image classification unlearning benchmarks show that OPC achieves not only effective unlearning performance but also superior resilience against both performance recovery attack and gradient-inversion attack. The distinctive unlearning performance of OPC arises from the deep feature forgetting enforced by its theoretical foundation, and recaps the need for improved robustness of machine unlearning methods.

Jaeheun Jung, Jaehyuk Lee, Yeajin Lee et al.

With the growth of demand on neural network compression methods, the structured pruning methods including importance-based approach are actively studied. The magnitude importance and many correlated modern importance criteria often limit the capacity of pruning decision, since the filters with larger magnitudes are not likely to be pruned if the smaller one didn't, even if it is redundant. In this paper, we propose a novel pruning strategy to challenge this dominating effect of magnitude and provide fair chance to each filter to be pruned, by placing it on projective space. After that, we observe the gradient descent movement whether the filters move toward the origin or not, to measure how the filter is likely to be pruned. This measurement is used to construct PROscore, a novel importance score for IPPRO, a novel importance-based structured pruning with magnitude-indifference. Our evaluation results shows that the proposed importance criteria using the projective space achieves near-lossless pruning by reducing the performance drop in pruning, with promising performance after the finetuning. Our work debunks the ``size-matters'' myth in pruning and expands the frontier of importance-based pruning both theoretically and empirically.

Jaeheun Jung, Donghun Lee

Structured pruning aims to reduce the size and computational cost of deep neural networks by removing entire filters or channels. The traditional regularizers such as L1 or Group Lasso and its variants lead to magnitude-biased pruning decisions, such that the filters with small magnitudes are likely to be pruned. Also, they often entail pruning results with almost zero margin around pruning decision boundary, such that tiny perturbation in a filter magnitude can flip the pruning decision. In this paper, we identify the precise algebraic condition under which pruning operations preserve model performance, and use the condition to construct a novel regularizer defined in an extended parameter space via auxiliary catalyst variables. The proposed Catalyst regularization ensures fair pruning chance for each filters with theoretically provable zero bias to their magnitude and robust pruning behavior achieved by wide-margin bifurcation of magnitudes between the preserved and the pruned filters. The theoretical properties naturally lead to real-world effectiveness, as shown by empirical validations of Catalyst Pruning algorithm. Pruning results on various datasets and models are superior to state-of-the-art filter pruning methods, and at the same time confirm the predicted robust and fair pruning characteristics of Catalyst pruning.

Wanshan Cui, Yejin Jeong, Inwook Song et al.

Accurate prediction of polymer material properties through data-driven approaches greatly accelerates novel material development by reducing redundant experiments and trial-and-error processes. However, inevitable outliers in empirical measurements can severely skew machine learning results, leading to erroneous prediction models and suboptimal material designs. To address this limitation, we propose a novel approach to enhance dataset quality efficiently by integrating multi-algorithm outlier detection with selective re-experimentation of unreliable outlier cases. To validate the empirical effectiveness of the approach, we systematically construct a new dataset containing 701 measurements of three key mechanical properties: glass transition temperature ($T_g$), tan $δ$ peak, and crosslinking density ($v_{c}$). To demonstrate its general applicability, we report the performance improvements across multiple machine learning models, including Elastic Net, SVR, Random Forest, and TPOT, to predict the three key properties. Our method reliably reduces prediction error (RMSE) and significantly improves accuracy with minimal additional experimental work, requiring only about 5% of the dataset to be re-measured. These findings highlight the importance of data quality enhancement in achieving reliable machine learning applications in polymer science and present a scalable strategy for improving predictive reliability in materials science.

Jaeheun Jung, Jaehyuk Lee, Changhae Jung et al.

Shock waves caused by earthquakes can be devastating. Generating realistic earthquake-caused ground motion waveforms help reducing losses in lives and properties, yet generative models for the task tend to generate subpar waveforms. We present High-fidelity Earthquake Groundmotion Generation System (HEGGS) and demonstrate its superior performance using earthquakes from North American, East Asian, and European regions. HEGGS exploits the intrinsic characteristics of earthquake dataset and learns the waveforms using an end-to-end differentiable generator containing conditional latent diffusion model and hi-fidelity waveform construction model. We show the learning efficiency of HEGGS by training it on a single GPU machine and validate its performance using earthquake databases from North America, East Asia, and Europe, using diverse criteria from waveform generation tasks and seismology. Once trained, HEGGS can generate three dimensional E-N-Z seismic waveforms with accurate P/S phase arrivals, envelope correlation, signal-to-noise ratio, GMPE analysis, frequency content analysis, and section plot analysis.

Donghun Lee, Srivatsan Srinivasan, Finale Doshi-Velez

We introduce a novel apprenticeship learning algorithm to learn an expert's underlying reward structure in off-policy model-free \emph{batch} settings. Unlike existing methods that require a dynamics model or additional data acquisition for on-policy evaluation, our algorithm requires only the batch data of observed expert behavior. Such settings are common in real-world tasks---health care, finance or industrial processes ---where accurate simulators do not exist or data acquisition is costly. To address challenges in batch settings, we introduce Deep Successor Feature Networks(DSFN) that estimate feature expectations in an off-policy setting and a transition-regularized imitation network that produces a near-expert initial policy and an efficient feature representation. Our algorithm achieves superior results in batch settings on both control benchmarks and a vital clinical task of sepsis management in the Intensive Care Unit.

Omer Gottesman, Fredrik Johansson, Joshua Meier et al.

Much attention has been devoted recently to the development of machine learning algorithms with the goal of improving treatment policies in healthcare. Reinforcement learning (RL) is a sub-field within machine learning that is concerned with learning how to make sequences of decisions so as to optimize long-term effects. Already, RL algorithms have been proposed to identify decision-making strategies for mechanical ventilation, sepsis management and treatment of schizophrenia. However, before implementing treatment policies learned by black-box algorithms in high-stakes clinical decision problems, special care must be taken in the evaluation of these policies. In this document, our goal is to expose some of the subtleties associated with evaluating RL algorithms in healthcare. We aim to provide a conceptual starting point for clinical and computational researchers to ask the right questions when designing and evaluating algorithms for new ways of treating patients. In the following, we describe how choices about how to summarize a history, variance of statistical estimators, and confounders in more ad-hoc measures can result in unreliable, even misleading estimates of the quality of a treatment policy. We also provide suggestions for mitigating these effects---for while there is much promise for mining observational health data to uncover better treatment policies, evaluation must be performed thoughtfully.

Jaekoo Lee, Byunghan Lee, Jongyoon Song et al.

With a new era of cloud and big data, Database Management Systems (DBMSs) have become more crucial in numerous enterprise business applications in all the industries. Accordingly, the importance of their proactive and preventive maintenance has also increased. However, detecting problems by predefined rules or stochastic modeling has limitations, particularly when analyzing the data on high-dimensional Key Performance Indicators (KPIs) from a DBMS. In recent years, Deep Learning (DL) has opened new opportunities for this complex analysis. In this paper, we present two complementary DL approaches to detect anomalies in SAP HANA. A temporal learning approach is used to detect abnormal patterns based on unlabeled historical data, whereas a spatial learning approach is used to classify known anomalies based on labeled data. We implement a system in SAP HANA integrated with Google TensorFlow. The experimental results with real-world data confirm the effectiveness of the system and models.

Junha Lee, Donghun Lee, Dae-Kyoo Kim et al.

Cohesion is a core design quality that has a great impact on posterior development and maintenance. By the nature of software, the cohesion of a system is diminished as the system evolves. God classes are code defects resulting from software evolution, having heterogeneous responsibilities highly coupled with other classes and often large in size, which makes it difficult to maintain the system. The existing work on identifying and decomposing God classes heavily relies on internal class information to identify God classes and responsibilities. However, in object-oriented systems, responsibilities should be analyzed with respect to not only internal class information, but also method interactions. In this paper, we present a novel approach for detecting God classes and decomposing their responsibilities based on the semantics of methods and method interactions. We evaluate the approach using JMeter v2.5.1 and the results are promising.