Sunwoo Kim

Sunwoo Kim, Youngjo Min, Younghun Jung et al. · nvidia, utoronto

We propose a controllable style transfer framework based on Implicit Neural Representation that pixel-wisely controls the stylized output via test-time training. Unlike traditional image optimization methods that often suffer from unstable convergence and learning-based methods that require intensive training and have limited generalization ability, we present a model optimization framework that optimizes the neural networks during test-time with explicit loss functions for style transfer. After being test-time trained once, thanks to the flexibility of the INR-based model, our framework can precisely control the stylized images in a pixel-wise manner and freely adjust image resolution without further optimization or training. We demonstrate several applications.

Jaehyuk Heo, YongGi Jeong, Sunwoo Kim et al.

We describe an approach used in the Generic Boundary Event Captioning challenge at the Long-Form Video Understanding Workshop held at CVPR 2022. We designed a Rich Encoder-decoder framework for Video Event CAptioner (REVECA) that utilizes spatial and temporal information from the video to generate a caption for the corresponding the event boundary. REVECA uses frame position embedding to incorporate information before and after the event boundary. Furthermore, it employs features extracted using the temporal segment network and temporal-based pairwise difference method to learn temporal information. A semantic segmentation mask for the attentional pooling process is adopted to learn the subject of an event. Finally, LoRA is applied to fine-tune the image encoder to enhance the learning efficiency. REVECA yielded an average score of 50.97 on the Kinetics-GEBC test data, which is an improvement of 10.17 over the baseline method. Our code is available in https://github.com/TooTouch/REVECA.

Jihye Park, Sunwoo Kim, Soohyun Kim et al. · nvidia, utoronto

Existing techniques for image-to-image translation commonly have suffered from two critical problems: heavy reliance on per-sample domain annotation and/or inability of handling multiple attributes per image. Recent truly-unsupervised methods adopt clustering approaches to easily provide per-sample one-hot domain labels. However, they cannot account for the real-world setting: one sample may have multiple attributes. In addition, the semantics of the clusters are not easily coupled to the human understanding. To overcome these, we present a LANguage-driven Image-to-image Translation model, dubbed LANIT. We leverage easy-to-obtain candidate attributes given in texts for a dataset: the similarity between images and attributes indicates per-sample domain labels. This formulation naturally enables multi-hot label so that users can specify the target domain with a set of attributes in language. To account for the case that the initial prompts are inaccurate, we also present prompt learning. We further present domain regularization loss that enforces translated images be mapped to the corresponding domain. Experiments on several standard benchmarks demonstrate that LANIT achieves comparable or superior performance to existing models.

Sunwoo Kim, Wooseok Jang, Hyunsu Kim et al. · nvidia, utoronto

Recent text-driven image editing in diffusion models has shown remarkable success. However, the existing methods assume that the user's description sufficiently grounds the contexts in the source image, such as objects, background, style, and their relations. This assumption is unsuitable for real-world applications because users have to manually engineer text prompts to find optimal descriptions for different images. From the users' standpoint, prompt engineering is a labor-intensive process, and users prefer to provide a target word for editing instead of a full sentence. To address this problem, we first demonstrate the importance of a detailed text description of the source image, by dividing prompts into three categories based on the level of semantic details. Then, we propose simple yet effective methods by combining prompt generation frameworks, thereby making the prompt engineering process more user-friendly. Extensive qualitative and quantitative experiments demonstrate the importance of prompts in text-driven image editing and our method is comparable to ground-truth prompts.

Dongyeong Hwang, Hyunju Kim, Sunwoo Kim et al.

The success of a specific neural network architecture is closely tied to the dataset and task it tackles; there is no one-size-fits-all solution. Thus, considerable efforts have been made to quickly and accurately estimate the performances of neural architectures, without full training or evaluation, for given tasks and datasets. Neural architecture encoding has played a crucial role in the estimation, and graphbased methods, which treat an architecture as a graph, have shown prominent performance. For enhanced representation learning of neural architectures, we introduce FlowerFormer, a powerful graph transformer that incorporates the information flows within a neural architecture. FlowerFormer consists of two key components: (a) bidirectional asynchronous message passing, inspired by the flows; (b) global attention built on flow-based masking. Our extensive experiments demonstrate the superiority of FlowerFormer over existing neural encoding methods, and its effectiveness extends beyond computer vision models to include graph neural networks and auto speech recognition models. Our code is available at http://github.com/y0ngjaenius/CVPR2024_FLOWERFormer.

Sanghyun Lee, Sunwoo Kim, Seungryong Kim et al.

Test-time scaling through reward-guided generation remains largely unexplored for discrete diffusion models despite its potential as a promising alternative. In this work, we introduce Iterative Reward-Guided Refinement (IterRef), a novel test-time scaling method tailored to discrete diffusion that leverages reward-guided noising-denoising transitions to progressively refine misaligned intermediate states. We formalize this process within a Multiple-Try Metropolis (MTM) framework, proving convergence to the reward-aligned distribution. Unlike prior methods that assume the current state is already aligned with the reward distribution and only guide the subsequent transition, our approach explicitly refines each state in situ, progressively steering it toward the optimal intermediate distribution. Across both text and image domains, we evaluate IterRef on diverse discrete diffusion models and observe consistent improvements in reward-guided generation quality. In particular, IterRef achieves striking gains under low compute budgets, far surpassing prior state-of-the-art baselines.

Seungnyun Kim, Jihoon Moon, Jinhong Kim et al.

Recently, we are witnessing the remarkable progress and widespread adoption of sensing technologies in autonomous driving, robotics, and metaverse. Considering the rapid advancement of computer vision (CV) technology to analyze the sensing information, we anticipate a proliferation of wireless applications exploiting the sensing and CV technologies in 6G. In this article, we provide a holistic overview of the sensing and CV-aided wireless communications (SVWC) framework for 6G. By analyzing the high-resolution sensing information through the powerful CV techniques, SVWC can quickly and accurately understand the wireless environments and then perform the wireless tasks. To demonstrate the efficacy of SVWC, we design the whole process of SVWC including the sensing dataset collection, DL model training, and execution of realistic wireless tasks. From the numerical evaluations on 6G communication scenarios, we show that SVWC achieves considerable performance gains over the conventional 5G systems in terms of positioning accuracy, data rate, and access latency.

Sunwoo Kim, Shinhwan Kang, Fanchen Bu et al.

Hypergraphs are marked by complex topology, expressing higher-order interactions among multiple nodes with hyperedges, and better capturing the topology is essential for effective representation learning. Recent advances in generative self-supervised learning (SSL) suggest that hypergraph neural networks learned from generative self supervision have the potential to effectively encode the complex hypergraph topology. Designing a generative SSL strategy for hypergraphs, however, is not straightforward. Questions remain with regard to its generative SSL task, connection to downstream tasks, and empirical properties of learned representations. In light of the promises and challenges, we propose a novel generative SSL strategy for hypergraphs. We first formulate a generative SSL task on hypergraphs, hyperedge filling, and highlight its theoretical connection to node classification. Based on the generative SSL task, we propose a hypergraph SSL method, HypeBoy. HypeBoy learns effective general-purpose hypergraph representations, outperforming 16 baseline methods across 11 benchmark datasets.

Sunwoo Kim, Soo Yong Lee, Fanchen Bu et al.

Graph autoencoders (Graph-AEs) learn representations of given graphs by aiming to accurately reconstruct them. A notable application of Graph-AEs is graph-level anomaly detection (GLAD), whose objective is to identify graphs with anomalous topological structures and/or node features compared to the majority of the graph population. Graph-AEs for GLAD regard a graph with a high mean reconstruction error (i.e. mean of errors from all node pairs and/or nodes) as anomalies. Namely, the methods rest on the assumption that they would better reconstruct graphs with similar characteristics to the majority. We, however, report non-trivial counter-examples, a phenomenon we call reconstruction flip, and highlight the limitations of the existing Graph-AE-based GLAD methods. Specifically, we empirically and theoretically investigate when this assumption holds and when it fails. Through our analyses, we further argue that, while the reconstruction errors for a given graph are effective features for GLAD, leveraging the multifaceted summaries of the reconstruction errors, beyond just mean, can further strengthen the features. Thus, we propose a novel and simple GLAD method, named MUSE. The key innovation of MUSE involves taking multifaceted summaries of reconstruction errors as graph features for GLAD. This surprisingly simple method obtains SOTA performance in GLAD, performing best overall among 14 methods across 10 datasets.

Jongha Lee, Sunwoo Kim, Kijung Shin

To detect anomalies in real-world graphs, such as social, email, and financial networks, various approaches have been developed. While they typically assume static input graphs, most real-world graphs grow over time, naturally represented as edge streams. In this context, we aim to achieve three goals: (a) instantly detecting anomalies as they occur, (b) adapting to dynamically changing states, and (c) handling the scarcity of dynamic anomaly labels. In this paper, we propose SLADE (Self-supervised Learning for Anomaly Detection in Edge Streams) for rapid detection of dynamic anomalies in edge streams, without relying on labels. SLADE detects the shifts of nodes into abnormal states by observing deviations in their interaction patterns over time. To this end, it trains a deep neural network to perform two self-supervised tasks: (a) minimizing drift in node representations and (b) generating long-term interaction patterns from short-term ones. Failure in these tasks for a node signals its deviation from the norm. Notably, the neural network and tasks are carefully designed so that all required operations can be performed in constant time (w.r.t. the graph size) in response to each new edge in the input stream. In dynamic anomaly detection across four real-world datasets, SLADE outperforms nine competing methods, even those leveraging label supervision.

Soo Yong Lee, Sunwoo Kim, Fanchen Bu et al.

How would randomly shuffling feature vectors among nodes from the same class affect graph neural networks (GNNs)? The feature shuffle, intuitively, perturbs the dependence between graph topology and features (A-X dependence) for GNNs to learn from. Surprisingly, we observe a consistent and significant improvement in GNN performance following the feature shuffle. Having overlooked the impact of A-X dependence on GNNs, the prior literature does not provide a satisfactory understanding of the phenomenon. Thus, we raise two research questions. First, how should A-X dependence be measured, while controlling for potential confounds? Second, how does A-X dependence affect GNNs? In response, we (i) propose a principled measure for A-X dependence, (ii) design a random graph model that controls A-X dependence, (iii) establish a theory on how A-X dependence relates to graph convolution, and (iv) present empirical analysis on real-world graphs that align with the theory. We conclude that A-X dependence mediates the effect of graph convolution, such that smaller dependence improves GNN-based node classification.

Sunwoo Kim, Haneul Yoo, Alice Oh

Understanding how large language models (LLMs) internally represent and process their predictions is central to detecting uncertainty and preventing hallucinations. While several studies have shown that models encode uncertainty in their hidden states, it is underexplored how this affects the way they process such hidden states. In this work, we demonstrate that the dynamics of output token probabilities across layers for certain and uncertain outputs are largely aligned, revealing that uncertainty does not seem to affect inference dynamics. Specifically, we use the Tuned Lens, a variant of the Logit Lens, to analyze the layer-wise probability trajectories of final prediction tokens across 11 datasets and 5 models. Using incorrect predictions as those with higher epistemic uncertainty, our results show aligned trajectories for certain and uncertain predictions that both observe abrupt increases in confidence at similar layers. We balance this finding by showing evidence that more competent models may learn to process uncertainty differently. Our findings challenge the feasibility of leveraging simplistic methods for detecting uncertainty at inference. More broadly, our work demonstrates how interpretability methods may be used to investigate the way uncertainty affects inference.

Risha Surana, Adrian Law, Sunwoo Kim et al.

Clinical notes are often stored in unstructured or semi-structured formats after extraction from electronic medical record (EMR) systems, which complicates their use for secondary analysis and downstream clinical applications. Reliable identification of section boundaries is a key step toward structuring these notes, as sections such as history of present illness, medications, and discharge instructions each provide distinct clinical contexts. In this work, we evaluate rule-based baselines, domain-specific transformer models, and large language models for clinical note segmentation using a curated dataset of 1,000 notes from MIMIC-IV. Our experiments show that large API-based models achieve the best overall performance, with GPT-5-mini reaching a best average F1 of 72.4 across sentence-level and freetext segmentation. Lightweight baselines remain competitive on structured sentence-level tasks but falter on unstructured freetext. Our results provide guidance for method selection and lay the groundwork for downstream tasks such as information extraction, cohort identification, and automated summarization.

Sunwoo Kim, Hyunjin Hwang, Kijung Shin

The performance of a deep learning model on a specific task and dataset depends heavily on its neural architecture, motivating considerable efforts to rapidly and accurately identify architectures suited to the target task and dataset. To achieve this, researchers use machine learning models-typically neural architecture encoders-to predict the performance of a neural architecture. Many state-of-the-art encoders aim to capture information flow within a neural architecture, which reflects how information moves through the forward pass and backpropagation, via a specialized model structure. However, due to their complicated structures, these flow-based encoders are significantly slower to process neural architectures compared to simpler encoders, presenting a notable practical challenge. To address this, we propose FGP, a novel pre-training method for neural architecture encoding that trains an encoder to capture the information flow without requiring specialized model structures. FGP trains an encoder to reconstruct a flow surrogate, our proposed representation of the neural architecture's information flow. Our experiments show that FGP boosts encoder performance by up to 106% in Precision-1%, compared to the same encoder trained solely with supervised learning.

Jisu Nam, Gyuseong Lee, Sunwoo Kim et al.

The objective for establishing dense correspondence between paired images consists of two terms: a data term and a prior term. While conventional techniques focused on defining hand-designed prior terms, which are difficult to formulate, recent approaches have focused on learning the data term with deep neural networks without explicitly modeling the prior, assuming that the model itself has the capacity to learn an optimal prior from a large-scale dataset. The performance improvement was obvious, however, they often fail to address inherent ambiguities of matching, such as textureless regions, repetitive patterns, and large displacements. To address this, we propose DiffMatch, a novel conditional diffusion-based framework designed to explicitly model both the data and prior terms. Unlike previous approaches, this is accomplished by leveraging a conditional denoising diffusion model. DiffMatch consists of two main components: conditional denoising diffusion module and cost injection module. We stabilize the training process and reduce memory usage with a stage-wise training strategy. Furthermore, to boost performance, we introduce an inference technique that finds a better path to the accurate matching field. Our experimental results demonstrate significant performance improvements of our method over existing approaches, and the ablation studies validate our design choices along with the effectiveness of each component. Project page is available at https://ku-cvlab.github.io/DiffMatch/.

Sunwoo Kim, Soohyun Kim, Seungryong Kim

Recent techniques to solve photorealistic style transfer within deep convolutional neural networks (CNNs) generally require intensive training from large-scale datasets, thus having limited applicability and poor generalization ability to unseen images or styles. To overcome this, we propose a novel framework, dubbed Deep Translation Prior (DTP), to accomplish photorealistic style transfer through test-time training on given input image pair with untrained networks, which learns an image pair-specific translation prior and thus yields better performance and generalization. Tailored for such test-time training for style transfer, we present novel network architectures, with two sub-modules of correspondence and generation modules, and loss functions consisting of contrastive content, style, and cycle consistency losses. Our framework does not require offline training phase for style transfer, which has been one of the main challenges in existing methods, but the networks are to be solely learned during test-time. Experimental results prove that our framework has a better generalization ability to unseen image pairs and even outperforms the state-of-the-art methods.

Sunwoo Kim, Minje Kim

In this paper, we present a blockwise optimization method for masking-based networks (BLOOM-Net) for training scalable speech enhancement networks. Here, we design our network with a residual learning scheme and train the internal separator blocks sequentially to obtain a scalable masking-based deep neural network for speech enhancement. Its scalability lets it dynamically adjust the run-time complexity depending on the test time environment. To this end, we modularize our models in that they can flexibly accommodate varying needs for enhancement performance and constraints on the resources, incurring minimal memory or training overhead due to the added scalability. Our experiments on speech enhancement demonstrate that the proposed blockwise optimization method achieves the desired scalability with only a slight performance degradation compared to corresponding models trained end-to-end.

Peng Wu, Tales Imbiriba, Junha Park et al.

Localization and tracking of objects using data-driven methods is a popular topic due to the complexity in characterizing the physics of wireless channel propagation models. In these modeling approaches, data needs to be gathered to accurately train models, at the same time that user's privacy is maintained. An appealing scheme to cooperatively achieve these goals is known as Federated Learning (FL). A challenge in FL schemes is the presence of non-independent and identically distributed (non-IID) data, caused by unevenly exploration of different areas. In this paper, we consider the use of recent FL schemes to train a set of personalized models that are then optimally fused through Bayesian rules, which makes it appropriate in the context of indoor localization.

Sunwoo Kim, Minje Kim

In realistic speech enhancement settings for end-user devices, we often encounter only a few speakers and noise types that tend to reoccur in the specific acoustic environment. We propose a novel personalized speech enhancement method to adapt a compact denoising model to the test-time specificity. Our goal in this test-time adaptation is to utilize no clean speech target of the test speaker, thus fulfilling the requirement for zero-shot learning. To complement the lack of clean utterance, we employ the knowledge distillation framework. Instead of the missing clean utterance target, we distill the more advanced denoising results from an overly large teacher model, and use it as the pseudo target to train the small student model. This zero-shot learning procedure circumvents the process of collecting users' clean speech, a process that users are reluctant to comply due to privacy concerns and technical difficulty of recording clean voice. Experiments on various test-time conditions show that the proposed personalization method achieves significant performance gains compared to larger baseline networks trained from a large speaker- and noise-agnostic datasets. In addition, since the compact personalized models can outperform larger general-purpose models, we claim that the proposed method performs model compression with no loss of denoising performance.

Aswin Sivaraman, Sunwoo Kim, Minje Kim

Training personalized speech enhancement models is innately a no-shot learning problem due to privacy constraints and limited access to noise-free speech from the target user. If there is an abundance of unlabeled noisy speech from the test-time user, a personalized speech enhancement model can be trained using self-supervised learning. One straightforward approach to model personalization is to use the target speaker's noisy recordings as pseudo-sources. Then, a pseudo denoising model learns to remove injected training noises and recover the pseudo-sources. However, this approach is volatile as it depends on the quality of the pseudo-sources, which may be too noisy. As a remedy, we propose an improvement to the self-supervised approach through data purification. We first train an SNR predictor model to estimate the frame-by-frame SNR of the pseudo-sources. Then, the predictor's estimates are converted into weights which adjust the frame-by-frame contribution of the pseudo-sources towards training the personalized model. We empirically show that the proposed data purification step improves the usability of the speaker-specific noisy data in the context of personalized speech enhancement. Without relying on any clean speech recordings or speaker embeddings, our approach may be seen as privacy-preserving.

Sun Hong Lim, Sunwoo Kim, Byonghyo Shim et al.

In this paper, we propose a deep learning-based beam tracking method for millimeter-wave (mmWave)communications. Beam tracking is employed for transmitting the known symbols using the sounding beams and tracking time-varying channels to maintain a reliable communication link. When the pose of a user equipment (UE) device varies rapidly, the mmWave channels also tend to vary fast, which hinders seamless communication. Thus, models that can capture temporal behavior of mmWave channels caused by the motion of the device are required, to cope with this problem. Accordingly, we employa deep neural network to analyze the temporal structure and patterns underlying in the time-varying channels and the signals acquired by inertial sensors. We propose a model based on long short termmemory (LSTM) that predicts the distribution of the future channel behavior based on a sequence of input signals available at the UE. This channel distribution is used to 1) control the sounding beams adaptively for the future channel state and 2) update the channel estimate through the measurement update step under a sequential Bayesian estimation framework. Our experimental results demonstrate that the proposed method achieves a significant performance gain over the conventional beam tracking methods under various mobility scenarios.

Taewon Kang, Soohyun Kim, Sunwoo Kim et al.

Existing techniques to solve exemplar-based image-to-image translation within deep convolutional neural networks (CNNs) generally require a training phase to optimize the network parameters on domain-specific and task-specific benchmarks, thus having limited applicability and generalization ability. In this paper, we propose a novel framework, for the first time, to solve exemplar-based translation through an online optimization given an input image pair, called online exemplar fine-tuning (OEFT), in which we fine-tune the off-the-shelf and general-purpose networks to the input image pair themselves. We design two sub-networks, namely correspondence fine-tuning and multiple GAN inversion, and optimize these network parameters and latent codes, starting from the pre-trained ones, with well-defined loss functions. Our framework does not require the off-line training phase, which has been the main challenge of existing methods, but the pre-trained networks to enable optimization in online. Experimental results prove that our framework is effective in having a generalization power to unseen image pairs and clearly even outperforms the state-of-the-arts needing the intensive training phase.

Sunwoo Kim, Haici Yang, Minje Kim

Speech enhancement tasks have seen significant improvements with the advance of deep learning technology, but with the cost of increased computational complexity. In this study, we propose an adaptive boosting approach to learning locality sensitive hash codes, which represent audio spectra efficiently. We use the learned hash codes for single-channel speech denoising tasks as an alternative to a complex machine learning model, particularly to address the resource-constrained environments. Our adaptive boosting algorithm learns simple logistic regressors as the weak learners. Once trained, their binary classification results transform each spectrum of test noisy speech into a bit string. Simple bitwise operations calculate Hamming distance to find the K-nearest matching frames in the dictionary of training noisy speech spectra, whose associated ideal binary masks are averaged to estimate the denoising mask for that test mixture. Our proposed learning algorithm differs from AdaBoost in the sense that the projections are trained to minimize the distances between the self-similarity matrix of the hash codes and that of the original spectra, rather than the misclassification rate. We evaluate our discriminative hash codes on the TIMIT corpus with various noise types, and show comparative performance to deep learning methods in terms of denoising performance and complexity.

Sunwoo Kim, Minje Kim

We propose an iteration-free source separation algorithm based on Winner-Take-All (WTA) hash codes, which is a faster, yet accurate alternative to a complex machine learning model for single-channel source separation in a resource-constrained environment. We first generate random permutations with WTA hashing to encode the shape of the multidimensional audio spectrum to a reduced bitstring representation. A nearest neighbor search on the hash codes of an incoming noisy spectrum as the query string results in the closest matches among the hashed mixture spectra. Using the indices of the matching frames, we obtain the corresponding ideal binary mask vectors for denoising. Since both the training data and the search operation are bitwise, the procedure can be done efficiently in hardware implementations. Experimental results show that the WTA hash codes are discriminant and provide an affordable dictionary search mechanism that leads to a competent performance compared to a comprehensive model and oracle masking.

Sunwoo Kim, Mrinmoy Maity, Minje Kim

This paper proposes a Bitwise Gated Recurrent Unit (BGRU) network for the single-channel source separation task. Recurrent Neural Networks (RNN) require several sets of weights within its cells, which significantly increases the computational cost compared to the fully-connected networks. To mitigate this increased computation, we focus on the GRU cells and quantize the feedforward procedure with binarized values and bitwise operations. The BGRU network is trained in two stages. The real-valued weights are pretrained and transferred to the bitwise network, which are then incrementally binarized to minimize the potential loss that can occur from a sudden introduction of quantization. As the proposed binarization technique turns only a few randomly chosen parameters into their binary versions, it gives the network training procedure a chance to gently adapt to the partly quantized version of the network. It eventually achieves the full binarization by incrementally increasing the amount of binarization over the iterations. Our experiments show that the proposed BGRU method produces source separation results greater than that of a real-valued fully connected network, with 11-12 dB mean Signal-to-Distortion Ratio (SDR). A fully binarized BGRU still outperforms a Bitwise Neural Network (BNN) by 1-2 dB even with less number of layers.