Jinhee Kim

Dongwei Wang, Jinhee Kim, Seokho Han et al.

Dynamic runtime latency and memory constraints necessitate flexible large language model (LLM) deployment, where an LLM can be inferred with various quantization precisions based on available computational resources. Recent work on such any-precision quantization either relies on hardware-inefficient vector quantization or induces additional scaling factors when switching between bit-widths. Meanwhile, existing post-training quantization (PTQ) methods calibrated for a fixed low precision show poor generalizability under runtime precision change. In this work, we attribute the source of poor generalization across bit-widths to a precision-dependent \textit{outlier migration} phenomenon where the distribution of PTQ-sensitive tokens changes across precisions. Motivated by this observation, we propose \texttt{MoBiQuant}, a novel any-precision Mixture-of-Bits quantization framework that adjusts weight precision for flexible LLM inference based on token sensitivity. Specifically, we propose a many-in-one recursive residual quantization that can iteratively reconstruct higher-precision weights at runtime and mitigates \textit{outlier migration} with a token-aware router to dynamically select the optimal inference precision of each token.Extensive experiments show that \texttt{MoBiQuant} matches or surpasses frontier single-precision PTQ while exhibiting strong elasticity, achieving significant memory savings and throughput gains of up to $1.34\times$ over state-of-the-art any-precision methods.

Jinhee Kim, Taesung Kim, Taewoo Kim et al.

Diagnosis based on medical images, such as X-ray images, often involves manual annotation of anatomical keypoints. However, this process involves significant human efforts and can thus be a bottleneck in the diagnostic process. To fully automate this procedure, deep-learning-based methods have been widely proposed and have achieved high performance in detecting keypoints in medical images. However, these methods still have clinical limitations: accuracy cannot be guaranteed for all cases, and it is necessary for doctors to double-check all predictions of models. In response, we propose a novel deep neural network that, given an X-ray image, automatically detects and refines the anatomical keypoints through a user-interactive system in which doctors can fix mispredicted keypoints with fewer clicks than needed during manual revision. Using our own collected data and the publicly available AASCE dataset, we demonstrate the effectiveness of the proposed method in reducing the annotation costs via extensive quantitative and qualitative results. A demo video of our approach is available on our project webpage.

Eungbeom Kim, Jinhee Kim, Yoori Oh et al.

In this paper, we aim to unveil the impact of data augmentation in audio-language multi-modal learning, which has not been explored despite its importance. We explore various augmentation methods at not only train-time but also test-time and find out that proper data augmentation can lead to substantial improvements. Specifically, applying our proposed audio-language paired augmentation PairMix, which is the first multi-modal audio-language augmentation method, outperforms the baselines for both automated audio captioning and audio-text retrieval tasks. To fully take advantage of data augmentation, we also present multi-level test-time augmentation (Multi-TTA) for the test-time. We successfully incorporate the two proposed methods and uni-modal augmentations and achieve 47.5 SPIDEr on audio captioning, which is an 18.2% relative increase over the baseline. In audio-text retrieval, the proposed methods also show an improvement in performance as well.

Wenhao Wu, Zishan Shao, Kangning Cui et al.

SVD-based Low-rank compression reduces transformer parameters and nominal FLOPs, but these savings often translate poorly into real LLM serving speedups. We show that this gap is largely a runtime problem: factorized checkpoints fragment execution paths, and the resulting overhead differs substantially between prefill and autoregressive decode. We present FlashSVD v1.5, a unified inference runtime for serving SVD-compressed transformers. FlashSVD v1.5 maps diverse public SVD compression families to a common factorized representation and combines phase-specific kernels with dense-KV decode, packed MLP execution, and per-layer CUDA-graph replay to reorganize the low-rank serving path into a thin runtime. Across representative decoder-serving settings, FlashSVD v1.5 achieves up to 2.55x decode and 2.39x end-to-end speedup, and it attains 1.48x average decode and 1.44x average end-to-end speedup across multiple popular SVD compression families. These results suggest that practical low-rank acceleration requires runtime co-design, not compression algorithms alone. Our code is available at: https://github.com/Zishan-Shao/FlashSVD.

Yukyeong Song, Jinhee Kim, Wanli Xing et al.

While mathematical creative writing can potentially engage students in expressing mathematical ideas in an imaginative way, some elementary school-age students struggle in this process. Generative AI (GenAI) offers possibilities for supporting creative writing activities, such as providing story generation. However, the design of GenAI-powered learning technologies requires careful consideration of the technology reception in the actual classrooms. This study explores students' and teachers' perceptions of creative mathematical writing with the developed GenAI-powered technology. The study adopted a qualitative thematic analysis of the interviews, triangulated with open-ended survey responses and classroom observation of 79 elementary school students, resulting in six themes and 19 subthemes. This study contributes by investigating the lived experience of GenAI-supported learning and the design considerations for GenAI-powered learning technologies and instructions.

Yukyeong Song, Jinhee Kim, Zifeng Liu et al.

Ongoing advancements in Generative AI (GenAI) have boosted the potential of applying long-standing learning-by-teaching practices in the form of a teachable agent (TA). Despite the recognized roles and opportunities of TAs, less is known about how GenAI could create synergy or introduce challenges in TAs and how students perceived the application of GenAI in TAs. This study explored middle school students perceived roles, benefits, and challenges of GenAI-powered TAs in an authentic mathematics classroom. Through classroom observation, focus-group interviews, and open-ended surveys of 108 sixth-grade students, we found that students expected the GenAI-powered TA to serve as a learning companion, facilitator, and collaborative problem-solver. Students also expressed the benefits and challenges of GenAI-powered TAs. This study provides implications for the design of educational AI and AI-assisted instruction.

Jinhee Kim, Taesung Kim, Jaegul Choo

Recent advances in interactive keypoint estimation methods have enhanced accuracy while minimizing user intervention. However, these methods require user input for error correction, which can be costly in vertebrae keypoint estimation where inaccurate keypoints are densely clustered or overlap. We introduce a novel approach, KeyBot, specifically designed to identify and correct significant and typical errors in existing models, akin to user revision. By characterizing typical error types and using simulated errors for training, KeyBot effectively corrects these errors and significantly reduces user workload. Comprehensive quantitative and qualitative evaluations on three public datasets confirm that KeyBot significantly outperforms existing methods, achieving state-of-the-art performance in interactive vertebrae keypoint estimation. The source code and demo video are available at: https://ts-kim.github.io/KeyBot/

Qinsi Wang, Hancheng Ye, Jinhee Kim et al.

Think about how human handles complex reading tasks: marking key points, inferring their relationships, and structuring information to guide understanding and responses. Likewise, can a large language model benefit from text structure to enhance text-processing performance? To explore it, in this work, we first introduce Structure of Thought (SoT), a prompting technique that explicitly guides models to construct intermediate text structures, consistently boosting performance across eight tasks and three model families. Building upon this insight, we present T2S-Bench, the first benchmark designed to evaluate and improve text-to-structure capabilities of models. T2S-Bench includes 1.8K samples across 6 scientific domains and 32 structural types, rigorously constructed to ensure accuracy, fairness, and quality. Evaluation on 45 mainstream models reveals substantial improvement potential: the average accuracy on the multi-hop reasoning task is only 52.1%, and even the most advanced model achieves 58.1% node accuracy in end-to-end extraction. Furthermore, on Qwen2.5-7B-Instruct, SoT alone yields an average +5.7% improvement across eight diverse text-processing tasks, and fine-tuning on T2S-Bench further increases this gain to +8.6%. These results highlight the value of explicit text structuring and the complementary contributions of SoT and T2S-Bench. Dataset and eval code have been released at https://t2s-bench.github.io/T2S-Bench-Page/.

Jinhee Kim, Jae Jun An, Kang Eun Jeon et al.

Multi-bit quantization networks enable flexible deployment of deep neural networks by supporting multiple precision levels within a single model. However, existing approaches suffer from significant training overhead as full-dataset updates are repeated for each supported bit-width, resulting in a cost that scales linearly with the number of precisions. Additionally, extra fine-tuning stages are often required to support additional or intermediate precision options, further compounding the overall training burden. To address this issue, we propose two techniques that greatly reduce the training overhead without compromising model utility: (i) Weight bias correction enables shared batch normalization and eliminates the need for fine-tuning by neutralizing quantization-induced bias across bit-widths and aligning activation distributions; and (ii) Bit-wise coreset sampling strategy allows each child model to train on a compact, informative subset selected via gradient-based importance scores by exploiting the implicit knowledge transfer phenomenon. Experiments on CIFAR-10/100, TinyImageNet, and ImageNet-1K with both ResNet and ViT architectures demonstrate that our method achieves competitive or superior accuracy while reducing training time up to 7.88x. Our code is released at https://github.com/a2jinhee/EMQNet_jk.

Jinhee Kim, Seoyeon Yoon, Taeho Lee et al.

The deployment of deep neural networks on edge devices is a challenging task due to the increasing complexity of state-of-the-art models, requiring efforts to reduce model size and inference latency. Recent studies explore models operating at diverse quantization settings to find the optimal point that balances computational efficiency and accuracy. Truncation, an effective approach for achieving lower bit precision mapping, enables a single model to adapt to various hardware platforms with little to no cost. However, formulating a training scheme for deep neural networks to withstand the associated errors introduced by truncation remains a challenge, as the current quantization-aware training schemes are not designed for the truncation process. We propose TruncQuant, a novel truncation-ready training scheme allowing flexible bit precision through bit-shifting in runtime. We achieve this by aligning TruncQuant with the output of the truncation process, demonstrating strong robustness across bit-width settings, and offering an easily implementable training scheme within existing quantization-aware frameworks. Our code is released at https://github.com/a2jinhee/TruncQuant.

Jinhee Kim, Taesung Kim, Jaegul Choo

Large language models (LLMs) have demonstrated remarkable in-context learning capabilities across diverse applications. In this work, we explore the effectiveness of LLMs for generating realistic synthetic tabular data, identifying key prompt design elements to optimize performance. We introduce EPIC, a novel approach that leverages balanced, grouped data samples and consistent formatting with unique variable mapping to guide LLMs in generating accurate synthetic data across all classes, even for imbalanced datasets. Evaluations on real-world datasets show that EPIC achieves state-of-the-art machine learning classification performance, significantly improving generation efficiency. These findings highlight the effectiveness of EPIC for synthetic tabular data generation, particularly in addressing class imbalance. Our source code for our work is available at: https://seharanul17.github.io/project-synthetic-tabular-llm/

Kang Eun Jeon, Sangheum Yeon, Jinhee Kim et al.

This paper addresses two critical challenges in analog In-Memory Computing (IMC) systems that limit their scalability and deployability: the computational unreliability caused by stuck-at faults (SAFs) and the high compilation overhead of existing fault-mitigation algorithms, namely Fault-Free (FF). To overcome these limitations, we first propose a novel multi-bit weight representation technique, termed row-column hybrid grouping, which generalizes conventional column grouping by introducing redundancy across both rows and columns. This structural redundancy enhances fault tolerance and can be effectively combined with existing fault-mitigation solutions. Second, we design a compiler pipeline that reformulates the fault-aware weight decomposition problem as an Integer Linear Programming (ILP) task, enabling fast and scalable compilation through off-the-shelf solvers. Further acceleration is achieved through theoretical insights that identify fault patterns amenable to trivial solutions, significantly reducing computation. Experimental results on convolutional networks and small language models demonstrate the effectiveness of our approach, achieving up to 8%p improvement in accuracy, 150x faster compilation, and 2x energy efficiency gain compared to existing baselines.

Seokho Han, Seoyeon Yoon, Jinhee Kim et al.

As deep neural networks (DNNs) see increased deployment on mobile and edge devices, optimizing model efficiency has become crucial. Mixed-precision quantization is widely favored, as it offers a superior balance between efficiency and accuracy compared to uniform quantization. However, finding the optimal precision for each layer is challenging. Recent studies utilizing bit-level sparsity have shown promise, yet they often introduce substantial training complexity and high GPU memory requirements. In this paper, we propose Memory-Efficient Bit Sparsification Quantization (MSQ), a novel approach that addresses these limitations. MSQ applies a round-clamp quantizer to enable differentiable computation of the least significant bits (LSBs) from model weights. It further employs regularization to induce sparsity in these LSBs, enabling effective precision reduction without explicit bit-level parameter splitting. Additionally, MSQ incorporates Hessian information, allowing the simultaneous pruning of multiple LSBs to further enhance training efficiency. Experimental results show that MSQ achieves up to 8.00x reduction in trainable parameters and up to 86% reduction in training time compared to previous bit-level quantization, while maintaining competitive accuracy and compression rates. This makes it a practical solution for training efficient DNNs on resource-constrained devices.

Jinhee Kim, Taesung Kim, Taewoo Kim et al.

In pediatric orthodontics, accurate estimation of growth potential is essential for developing effective treatment strategies. Our research aims to predict this potential by identifying the growth peak and analyzing cervical vertebra morphology solely through lateral cephalometric radiographs. We accomplish this by comprehensively analyzing cervical vertebral maturation (CVM) features from these radiographs. This methodology provides clinicians with a reliable and efficient tool to determine the optimal timings for orthodontic interventions, ultimately enhancing patient outcomes. A crucial aspect of this approach is the meticulous annotation of keypoints on the cervical vertebrae, a task often challenged by its labor-intensive nature. To mitigate this, we introduce Attend-and-Refine Network (ARNet), a user-interactive, deep learning-based model designed to streamline the annotation process. ARNet features Interaction-guided recalibration network, which adaptively recalibrates image features in response to user feedback, coupled with a morphology-aware loss function that preserves the structural consistency of keypoints. This novel approach substantially reduces manual effort in keypoint identification, thereby enhancing the efficiency and accuracy of the process. Extensively validated across various datasets, ARNet demonstrates remarkable performance and exhibits wide-ranging applicability in medical imaging. In conclusion, our research offers an effective AI-assisted diagnostic tool for assessing growth potential in pediatric orthodontics, marking a significant advancement in the field.

Saemee Choi, Sohyun Jeong, Jaegul Choo et al.

We propose ImEdit, the first zero-shot, training-free video editing method conditioned on both images and text. The proposed method introduces $ρ$-start sampling and dilated dual masking to construct well-structured noise maps for coherent and accurate edits. We further present zero image guidance, a controllable negative prompt strategy, for visual fidelity. Both quantitative and qualitative evaluations show that our method outperforms state-of-the-art methods across all metrics.

Do Yeong Kang, Yeong Hwan Oh, Chanwook Hwang et al.

The implementation of Hyperdimensional Computing (HDC) on In-Memory Computing (IMC) architectures faces significant challenges due to the mismatch between highdimensional vectors and IMC array sizes, leading to inefficient memory utilization and increased computation cycles. This paper presents MEMHD, a Memory-Efficient Multi-centroid HDC framework designed to address these challenges. MEMHD introduces a clustering-based initialization method and quantization aware iterative learning for multi-centroid associative memory. Through these approaches and its overall architecture, MEMHD achieves a significant reduction in memory requirements while maintaining or improving classification accuracy. Our approach achieves full utilization of IMC arrays and enables one-shot (or few-shot) associative search. Experimental results demonstrate that MEMHD outperforms state-of-the-art binary HDC models, achieving up to 13.69% higher accuracy with the same memory usage, or 13.25x more memory efficiency at the same accuracy level. Moreover, MEMHD reduces computation cycles by up to 80x and array usage by up to 71x compared to baseline IMC mapping methods when mapped to 128x128 IMC arrays, while significantly improving energy and computation cycle efficiency.